diff --git a/README.md b/README.md index b294e54b..b35776dc 100644 --- a/README.md +++ b/README.md @@ -36,9 +36,13 @@ pip install fastNLP python -m spacy download en ``` +目前使用pip安装fastNLP的版本是0.4.1，有较多功能仍未更新，最新内容以master分支为准。 +fastNLP0.5.0版本将在近期推出，请密切关注。 + ## fastNLP教程 +- [0. 快速入门](https://fastnlp.readthedocs.io/zh/latest/user/quickstart.html) - [1. 使用DataSet预处理文本](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_1_data_preprocess.html) - [2. 使用DataSetLoader加载数据集](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_2_load_dataset.html) - [3. 使用Embedding模块将文本转成向量](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_3_embedding.html) @@ -48,17 +52,23 @@ python -m spacy download en - [7. 使用Modules和Models快速搭建自定义模型](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_7_modules_models.html) - [8. 使用Metric快速评测你的模型](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_8_metrics.html) - [9. 使用Callback自定义你的训练过程](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_9_callback.html) +- [10. 使用fitlog 辅助 fastNLP 进行科研](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_10_fitlog.html) ## 内置组件 -大部分用于的 NLP 任务神经网络都可以看做由编码器（encoder）、解码器（decoder）两种模块组成。 +大部分用于的 NLP 任务神经网络都可以看做由词嵌入（embeddings）和两种模块：编码器（encoder）、解码器（decoder）组成。 + +以文本分类任务为例，下图展示了一个BiLSTM+Attention实现文本分类器的模型流程图： ![](./docs/source/figures/text_classification.png) -fastNLP 在 modules 模块中内置了两种模块的诸多组件，可以帮助用户快速搭建自己所需的网络。 两种模块的功能和常见组件如下: +fastNLP 在 embeddings 模块中内置了几种不同的embedding：静态embedding（GloVe、word2vec）、上下文相关embedding +（ELMo、BERT）、字符embedding（基于CNN或者LSTM的CharEmbedding） + +与此同时，fastNLP 在 modules 模块中内置了两种模块的诸多组件，可以帮助用户快速搭建自己所需的网络。 两种模块的功能和常见组件如下: @@ -102,6 +112,10 @@ fastNLP的大致工作流程如上图所示，而项目结构如下： + + + + diff --git a/docs/Makefile b/docs/Makefile index 6ba2fa54..2b4de2d8 100644 --- a/docs/Makefile +++ b/docs/Makefile @@ -19,6 +19,9 @@ apidoc: server: cd build/html && python -m http.server +dev: + rm -rf build/html && make html && make server + .PHONY: help Makefile # Catch-all target: route all unknown targets to Sphinx using the new diff --git a/docs/README.md b/docs/README.md new file mode 100644 index 00000000..15dcccda --- /dev/null +++ b/docs/README.md @@ -0,0 +1,41 @@ +# 快速入门 fastNLP 文档编写 + +本教程为 fastNLP 文档编写者创建，文档编写者包括合作开发人员和文档维护人员。您在一般情况下属于前者， +只需要了解整个框架的部分内容即可。 + +## 合作开发人员 + +FastNLP的文档使用基于[reStructuredText标记语言](http://docutils.sourceforge.net/rst.html)的 +[Sphinx](http://sphinx.pocoo.org/)工具生成，由[Read the Docs](https://readthedocs.org/)网站自动维护生成。 +一般开发者只要编写符合reStructuredText语法规范的文档并通过[PR](https://help.github.com/en/articles/about-pull-requests)， +就可以为fastNLP的文档贡献一份力量。 + +如果你想在本地编译文档并进行大段文档的编写，您需要安装Sphinx工具以及sphinx-rtd-theme主题： +```bash +fastNLP/docs> pip install sphinx +fastNLP/docs> pip install sphinx-rtd-theme +``` +然后在本目录下执行 `make dev` 命令。该命令只支持Linux和MacOS系统，期望看到如下输出： +```bash +fastNLP/docs> make dev +rm -rf build/html && make html && make server +Running Sphinx v1.5.6 +making output directory... +...... +Build finished. The HTML pages are in build/html. +cd build/html && python -m http.server +Serving HTTP on 0.0.0.0 port 8000 (http://0.0.0.0:8000/) ... +``` +现在您浏览器访问 http://localhost:8000/ 查看文档。如果你在远程服务器尚进行工作，则访问地址为 http://{服务器的ip地址}:8000/ 。 +但您必须保证服务器的8000端口是开放的。如果您的电脑或远程服务器的8000端口被占用，程序会顺延使用8001、8002……等端口。 +当你结束访问时，您可以使用Control(Ctrl) + C 来结束进程。 + +我们在[这里](./source/user/example.rst)列举了fastNLP文档经常用到的reStructuredText语法（网页查看请结合Raw模式）， +您可以通过阅读它进行快速上手。FastNLP大部分的文档都是写在代码中通过Sphinx工具进行抽取生成的， +您还可以参考这篇[未完成的文章](./source/user/docs_in_code.rst)了解代码内文档编写的规范。 + +## 文档维护人员 + +文档维护人员需要了解 Makefile 中全部命令的含义，并了解到目前的文档结构 +是在 sphinx-apidoc 自动抽取的基础上进行手动修改得到的。 +文档维护人员应进一步提升整个框架的自动化程度，并监督合作开发人员不要破坏文档项目的整体结构。 \ No newline at end of file diff --git a/docs/make.bat b/docs/make.bat deleted file mode 100644 index 1c651b1f..00000000 --- a/docs/make.bat +++ /dev/null @@ -1,36 +0,0 @@ -@ECHO OFF - -pushd %~dp0 - -REM Command file for Sphinx documentation - -if "%SPHINXBUILD%" == "" ( - set SPHINXBUILD=sphinx-build -) -set SOURCEDIR=source -set BUILDDIR=build -set SPHINXPROJ=fastNLP - -if "%1" == "" goto help - -%SPHINXBUILD% >NUL 2>NUL -if errorlevel 9009 ( - echo. - echo.The 'sphinx-build' command was not found. Make sure you have Sphinx - echo.installed, then set the SPHINXBUILD environment variable to point - echo.to the full path of the 'sphinx-build' executable. Alternatively you - echo.may add the Sphinx directory to PATH. - echo. - echo.If you don't have Sphinx installed, grab it from - echo.http://sphinx-doc.org/ - exit /b 1 -) - -%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% -goto end - -:help -%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% - -:end -popd diff --git a/docs/quick_tutorial.md b/docs/quick_tutorial.md deleted file mode 100644 index 64c51124..00000000 --- a/docs/quick_tutorial.md +++ /dev/null @@ -1,2 +0,0 @@ -# FastNLP Quick Tutorial - diff --git a/docs/source/conf.py b/docs/source/conf.py index 3e9753af..2e10bc89 100644 --- a/docs/source/conf.py +++ b/docs/source/conf.py @@ -24,9 +24,9 @@ copyright = '2018, xpqiu' author = 'xpqiu' # The short X.Y version -version = '0.4' +version = '0.4.5' # The full version, including alpha/beta/rc tags -release = '0.4' +release = '0.4.5' # -- General configuration --------------------------------------------------- diff --git a/docs/source/fastNLP.core.batch.rst b/docs/source/fastNLP.core.batch.rst index 33a5b730..03008b52 100644 --- a/docs/source/fastNLP.core.batch.rst +++ b/docs/source/fastNLP.core.batch.rst @@ -2,6 +2,6 @@ fastNLP.core.batch ================== .. automodule:: fastNLP.core.batch - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.callback.rst b/docs/source/fastNLP.core.callback.rst index 31ec627b..74a7825d 100644 --- a/docs/source/fastNLP.core.callback.rst +++ b/docs/source/fastNLP.core.callback.rst @@ -2,6 +2,6 @@ fastNLP.core.callback ===================== .. automodule:: fastNLP.core.callback - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.const.rst b/docs/source/fastNLP.core.const.rst index c9e3bd97..330a8883 100644 --- a/docs/source/fastNLP.core.const.rst +++ b/docs/source/fastNLP.core.const.rst @@ -2,6 +2,6 @@ fastNLP.core.const ================== .. automodule:: fastNLP.core.const - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.dataset.rst b/docs/source/fastNLP.core.dataset.rst index b377cb0f..1ad94bb6 100644 --- a/docs/source/fastNLP.core.dataset.rst +++ b/docs/source/fastNLP.core.dataset.rst @@ -2,6 +2,6 @@ fastNLP.core.dataset ==================== .. automodule:: fastNLP.core.dataset - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.field.rst b/docs/source/fastNLP.core.field.rst index 7686e79a..7fc099c9 100644 --- a/docs/source/fastNLP.core.field.rst +++ b/docs/source/fastNLP.core.field.rst @@ -2,6 +2,6 @@ fastNLP.core.field ================== .. automodule:: fastNLP.core.field - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.instance.rst b/docs/source/fastNLP.core.instance.rst index 14393a91..6e496ac1 100644 --- a/docs/source/fastNLP.core.instance.rst +++ b/docs/source/fastNLP.core.instance.rst @@ -2,6 +2,6 @@ fastNLP.core.instance ===================== .. automodule:: fastNLP.core.instance - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.losses.rst b/docs/source/fastNLP.core.losses.rst index d2dd492b..8e63dfa1 100644 --- a/docs/source/fastNLP.core.losses.rst +++ b/docs/source/fastNLP.core.losses.rst @@ -2,6 +2,6 @@ fastNLP.core.losses =================== .. automodule:: fastNLP.core.losses - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.metrics.rst b/docs/source/fastNLP.core.metrics.rst index 69afff36..d3b87bb8 100644 --- a/docs/source/fastNLP.core.metrics.rst +++ b/docs/source/fastNLP.core.metrics.rst @@ -2,6 +2,6 @@ fastNLP.core.metrics ==================== .. automodule:: fastNLP.core.metrics - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.optimizer.rst b/docs/source/fastNLP.core.optimizer.rst index e2100d2e..c80be53f 100644 --- a/docs/source/fastNLP.core.optimizer.rst +++ b/docs/source/fastNLP.core.optimizer.rst @@ -2,6 +2,6 @@ fastNLP.core.optimizer ====================== .. automodule:: fastNLP.core.optimizer - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.rst b/docs/source/fastNLP.core.rst index 82c13e46..cacc6622 100644 --- a/docs/source/fastNLP.core.rst +++ b/docs/source/fastNLP.core.rst @@ -2,15 +2,15 @@ fastNLP.core ============ .. automodule:: fastNLP.core - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: 子模块 ---------- .. toctree:: - :titlesonly: + :maxdepth: 1 fastNLP.core.batch fastNLP.core.callback @@ -26,4 +26,3 @@ fastNLP.core fastNLP.core.trainer fastNLP.core.utils fastNLP.core.vocabulary - diff --git a/docs/source/fastNLP.core.sampler.rst b/docs/source/fastNLP.core.sampler.rst index 1810d59c..0110f0c0 100644 --- a/docs/source/fastNLP.core.sampler.rst +++ b/docs/source/fastNLP.core.sampler.rst @@ -2,6 +2,6 @@ fastNLP.core.sampler ==================== .. automodule:: fastNLP.core.sampler - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.tester.rst b/docs/source/fastNLP.core.tester.rst index a9e7e09f..4d71a27b 100644 --- a/docs/source/fastNLP.core.tester.rst +++ b/docs/source/fastNLP.core.tester.rst @@ -2,6 +2,6 @@ fastNLP.core.tester =================== .. automodule:: fastNLP.core.tester - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.trainer.rst b/docs/source/fastNLP.core.trainer.rst index 9e518d4b..60bf2d5b 100644 --- a/docs/source/fastNLP.core.trainer.rst +++ b/docs/source/fastNLP.core.trainer.rst @@ -2,6 +2,6 @@ fastNLP.core.trainer ==================== .. automodule:: fastNLP.core.trainer - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.utils.rst b/docs/source/fastNLP.core.utils.rst index fcd3f50c..3f80b4e8 100644 --- a/docs/source/fastNLP.core.utils.rst +++ b/docs/source/fastNLP.core.utils.rst @@ -2,6 +2,6 @@ fastNLP.core.utils ================== .. automodule:: fastNLP.core.utils - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.core.vocabulary.rst b/docs/source/fastNLP.core.vocabulary.rst index b3bf4bac..ba9598b9 100644 --- a/docs/source/fastNLP.core.vocabulary.rst +++ b/docs/source/fastNLP.core.vocabulary.rst @@ -2,6 +2,6 @@ fastNLP.core.vocabulary ======================= .. automodule:: fastNLP.core.vocabulary - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.bert_embedding.rst b/docs/source/fastNLP.embeddings.bert_embedding.rst new file mode 100644 index 00000000..24ceff1c --- /dev/null +++ b/docs/source/fastNLP.embeddings.bert_embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.bert\_embedding +================================== + +.. automodule:: fastNLP.embeddings.bert_embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.char_embedding.rst b/docs/source/fastNLP.embeddings.char_embedding.rst new file mode 100644 index 00000000..501089d8 --- /dev/null +++ b/docs/source/fastNLP.embeddings.char_embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.char\_embedding +================================== + +.. automodule:: fastNLP.embeddings.char_embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.elmo_embedding.rst b/docs/source/fastNLP.embeddings.elmo_embedding.rst new file mode 100644 index 00000000..76669ee3 --- /dev/null +++ b/docs/source/fastNLP.embeddings.elmo_embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.elmo\_embedding +================================== + +.. automodule:: fastNLP.embeddings.elmo_embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.embedding.rst b/docs/source/fastNLP.embeddings.embedding.rst new file mode 100644 index 00000000..5960d2cd --- /dev/null +++ b/docs/source/fastNLP.embeddings.embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.embedding +============================ + +.. automodule:: fastNLP.embeddings.embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.rst b/docs/source/fastNLP.embeddings.rst new file mode 100644 index 00000000..6b168906 --- /dev/null +++ b/docs/source/fastNLP.embeddings.rst @@ -0,0 +1,21 @@ +fastNLP.embeddings +================== + +.. automodule:: fastNLP.embeddings + :members: + :undoc-members: + :show-inheritance: + +子模块 +---------- + +.. toctree:: + :maxdepth: 1 + + fastNLP.embeddings.bert_embedding + fastNLP.embeddings.char_embedding + fastNLP.embeddings.elmo_embedding + fastNLP.embeddings.embedding + fastNLP.embeddings.stack_embedding + fastNLP.embeddings.static_embedding + fastNLP.embeddings.utils diff --git a/docs/source/fastNLP.embeddings.stack_embedding.rst b/docs/source/fastNLP.embeddings.stack_embedding.rst new file mode 100644 index 00000000..4d2115f7 --- /dev/null +++ b/docs/source/fastNLP.embeddings.stack_embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.stack\_embedding +=================================== + +.. automodule:: fastNLP.embeddings.stack_embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.static_embedding.rst b/docs/source/fastNLP.embeddings.static_embedding.rst new file mode 100644 index 00000000..e46de81a --- /dev/null +++ b/docs/source/fastNLP.embeddings.static_embedding.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.static\_embedding +==================================== + +.. automodule:: fastNLP.embeddings.static_embedding + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.embeddings.utils.rst b/docs/source/fastNLP.embeddings.utils.rst new file mode 100644 index 00000000..263bfbd6 --- /dev/null +++ b/docs/source/fastNLP.embeddings.utils.rst @@ -0,0 +1,7 @@ +fastNLP.embeddings.utils +======================== + +.. automodule:: fastNLP.embeddings.utils + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.io.base_loader.rst b/docs/source/fastNLP.io.base_loader.rst index c1f9ac14..057867f4 100644 --- a/docs/source/fastNLP.io.base_loader.rst +++ b/docs/source/fastNLP.io.base_loader.rst @@ -2,6 +2,6 @@ fastNLP.io.base\_loader ======================= .. automodule:: fastNLP.io.base_loader - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.io.data_loader.rst b/docs/source/fastNLP.io.data_loader.rst new file mode 100644 index 00000000..8f990102 --- /dev/null +++ b/docs/source/fastNLP.io.data_loader.rst @@ -0,0 +1,7 @@ +fastNLP.io.data\_loader +========================== + +.. automodule:: fastNLP.io.data_loader + :members: + :undoc-members: + :show-inheritance: \ No newline at end of file diff --git a/docs/source/fastNLP.io.dataset_loader.rst b/docs/source/fastNLP.io.dataset_loader.rst index d6663e59..e7990714 100644 --- a/docs/source/fastNLP.io.dataset_loader.rst +++ b/docs/source/fastNLP.io.dataset_loader.rst @@ -2,6 +2,6 @@ fastNLP.io.dataset\_loader ========================== .. automodule:: fastNLP.io.dataset_loader - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.io.embed_loader.rst b/docs/source/fastNLP.io.embed_loader.rst index 7a8e730c..69e1f7ff 100644 --- a/docs/source/fastNLP.io.embed_loader.rst +++ b/docs/source/fastNLP.io.embed_loader.rst @@ -2,6 +2,6 @@ fastNLP.io.embed\_loader ======================== .. automodule:: fastNLP.io.embed_loader - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.io.model_io.rst b/docs/source/fastNLP.io.model_io.rst index 50d4c25a..537ce752 100644 --- a/docs/source/fastNLP.io.model_io.rst +++ b/docs/source/fastNLP.io.model_io.rst @@ -2,6 +2,6 @@ fastNLP.io.model\_io ==================== .. automodule:: fastNLP.io.model_io - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.io.rst b/docs/source/fastNLP.io.rst index fad05a21..a97ed67d 100644 --- a/docs/source/fastNLP.io.rst +++ b/docs/source/fastNLP.io.rst @@ -2,18 +2,18 @@ fastNLP.io ========== .. automodule:: fastNLP.io - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: 子模块 ---------- .. toctree:: - :titlesonly: + :maxdepth: 1 fastNLP.io.base_loader - fastNLP.io.dataset_loader fastNLP.io.embed_loader + fastNLP.io.dataset_loader + fastNLP.io.data_loader fastNLP.io.model_io - diff --git a/docs/source/fastNLP.models.biaffine_parser.rst b/docs/source/fastNLP.models.biaffine_parser.rst index a3dd1836..f19504e8 100644 --- a/docs/source/fastNLP.models.biaffine_parser.rst +++ b/docs/source/fastNLP.models.biaffine_parser.rst @@ -2,6 +2,6 @@ fastNLP.models.biaffine\_parser =============================== .. automodule:: fastNLP.models.biaffine_parser - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.models.cnn_text_classification.rst b/docs/source/fastNLP.models.cnn_text_classification.rst index a935d0bf..eacf6916 100644 --- a/docs/source/fastNLP.models.cnn_text_classification.rst +++ b/docs/source/fastNLP.models.cnn_text_classification.rst @@ -2,6 +2,6 @@ fastNLP.models.cnn\_text\_classification ======================================== .. automodule:: fastNLP.models.cnn_text_classification - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.models.rst b/docs/source/fastNLP.models.rst index 5858ebcd..2ea546e2 100644 --- a/docs/source/fastNLP.models.rst +++ b/docs/source/fastNLP.models.rst @@ -2,19 +2,18 @@ fastNLP.models ============== .. automodule:: fastNLP.models - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: 子模块 ---------- .. toctree:: - :titlesonly: + :maxdepth: 1 fastNLP.models.biaffine_parser fastNLP.models.cnn_text_classification fastNLP.models.sequence_labeling fastNLP.models.snli fastNLP.models.star_transformer - diff --git a/docs/source/fastNLP.models.sequence_labeling.rst b/docs/source/fastNLP.models.sequence_labeling.rst index 6d569fe1..85e28f06 100644 --- a/docs/source/fastNLP.models.sequence_labeling.rst +++ b/docs/source/fastNLP.models.sequence_labeling.rst @@ -2,6 +2,6 @@ fastNLP.models.sequence\_labeling ================================= .. automodule:: fastNLP.models.sequence_labeling - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.models.snli.rst b/docs/source/fastNLP.models.snli.rst index 24c2cc53..3b9b555c 100644 --- a/docs/source/fastNLP.models.snli.rst +++ b/docs/source/fastNLP.models.snli.rst @@ -2,6 +2,6 @@ fastNLP.models.snli =================== .. automodule:: fastNLP.models.snli - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.models.star_transformer.rst b/docs/source/fastNLP.models.star_transformer.rst index c93fb8cd..69d5c5b2 100644 --- a/docs/source/fastNLP.models.star_transformer.rst +++ b/docs/source/fastNLP.models.star_transformer.rst @@ -2,6 +2,6 @@ fastNLP.models.star\_transformer ================================ .. automodule:: fastNLP.models.star_transformer - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.modules.aggregator.attention.rst b/docs/source/fastNLP.modules.aggregator.attention.rst deleted file mode 100644 index dc9c2b53..00000000 --- a/docs/source/fastNLP.modules.aggregator.attention.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.aggregator.attention -==================================== - -.. automodule:: fastNLP.modules.aggregator.attention - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.aggregator.pooling.rst b/docs/source/fastNLP.modules.aggregator.pooling.rst deleted file mode 100644 index 162f889d..00000000 --- a/docs/source/fastNLP.modules.aggregator.pooling.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.aggregator.pooling -================================== - -.. automodule:: fastNLP.modules.aggregator.pooling - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.aggregator.rst b/docs/source/fastNLP.modules.aggregator.rst deleted file mode 100644 index 44398325..00000000 --- a/docs/source/fastNLP.modules.aggregator.rst +++ /dev/null @@ -1,17 +0,0 @@ -fastNLP.modules.aggregator -========================== - -.. automodule:: fastNLP.modules.aggregator - :members: - :undoc-members: - :show-inheritance: - -子模块 ----------- - -.. toctree:: - :titlesonly: - - fastNLP.modules.aggregator.attention - fastNLP.modules.aggregator.pooling - diff --git a/docs/source/fastNLP.modules.decoder.crf.rst b/docs/source/fastNLP.modules.decoder.crf.rst deleted file mode 100644 index 6d5b0d5b..00000000 --- a/docs/source/fastNLP.modules.decoder.crf.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.decoder.CRF -=========================== - -.. automodule:: fastNLP.modules.decoder.crf - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.decoder.mlp.rst b/docs/source/fastNLP.modules.decoder.mlp.rst deleted file mode 100644 index 7d661ebf..00000000 --- a/docs/source/fastNLP.modules.decoder.mlp.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.decoder.MLP -=========================== - -.. automodule:: fastNLP.modules.decoder.mlp - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.decoder.rst b/docs/source/fastNLP.modules.decoder.rst index e42a9f39..ecc2adbd 100644 --- a/docs/source/fastNLP.modules.decoder.rst +++ b/docs/source/fastNLP.modules.decoder.rst @@ -2,17 +2,7 @@ fastNLP.modules.decoder ======================= .. automodule:: fastNLP.modules.decoder - :members: - :undoc-members: - :show-inheritance: - -子模块 ----------- - -.. toctree:: - :titlesonly: - - fastNLP.modules.decoder.crf - fastNLP.modules.decoder.mlp - fastNLP.modules.decoder.utils + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.modules.decoder.utils.rst b/docs/source/fastNLP.modules.decoder.utils.rst deleted file mode 100644 index da979d99..00000000 --- a/docs/source/fastNLP.modules.decoder.utils.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.decoder.utils -============================= - -.. automodule:: fastNLP.modules.decoder.utils - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.bert.rst b/docs/source/fastNLP.modules.encoder.bert.rst deleted file mode 100644 index 66bd0bbd..00000000 --- a/docs/source/fastNLP.modules.encoder.bert.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.bert -============================ - -.. automodule:: fastNLP.modules.encoder.bert - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.char_encoder.rst b/docs/source/fastNLP.modules.encoder.char_encoder.rst deleted file mode 100644 index 61ea3340..00000000 --- a/docs/source/fastNLP.modules.encoder.char_encoder.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.char\_encoder -===================================== - -.. automodule:: fastNLP.modules.encoder.char_encoder - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.conv_maxpool.rst b/docs/source/fastNLP.modules.encoder.conv_maxpool.rst deleted file mode 100644 index 7058a723..00000000 --- a/docs/source/fastNLP.modules.encoder.conv_maxpool.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.conv\_maxpool -===================================== - -.. automodule:: fastNLP.modules.encoder.conv_maxpool - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.embedding.rst b/docs/source/fastNLP.modules.encoder.embedding.rst deleted file mode 100644 index 4427b3bf..00000000 --- a/docs/source/fastNLP.modules.encoder.embedding.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.embedding -================================= - -.. automodule:: fastNLP.modules.encoder.embedding - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.lstm.rst b/docs/source/fastNLP.modules.encoder.lstm.rst deleted file mode 100644 index f9cbea88..00000000 --- a/docs/source/fastNLP.modules.encoder.lstm.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.lstm -============================ - -.. automodule:: fastNLP.modules.encoder.lstm - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.rst b/docs/source/fastNLP.modules.encoder.rst index b15232fa..0562f12d 100644 --- a/docs/source/fastNLP.modules.encoder.rst +++ b/docs/source/fastNLP.modules.encoder.rst @@ -2,22 +2,6 @@ fastNLP.modules.encoder ======================= .. automodule:: fastNLP.modules.encoder - :members: - :undoc-members: - :show-inheritance: - -子模块 ----------- - -.. toctree:: - :titlesonly: - - fastNLP.modules.encoder.bert - fastNLP.modules.encoder.char_encoder - fastNLP.modules.encoder.conv_maxpool - fastNLP.modules.encoder.embedding - fastNLP.modules.encoder.lstm - fastNLP.modules.encoder.star_transformer - fastNLP.modules.encoder.transformer - fastNLP.modules.encoder.variational_rnn - + :members: + :undoc-members: + :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.star_transformer.rst b/docs/source/fastNLP.modules.encoder.star_transformer.rst deleted file mode 100644 index 0c406782..00000000 --- a/docs/source/fastNLP.modules.encoder.star_transformer.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.star\_transformer -========================================= - -.. automodule:: fastNLP.modules.encoder.star_transformer - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.transformer.rst b/docs/source/fastNLP.modules.encoder.transformer.rst deleted file mode 100644 index 6a40c597..00000000 --- a/docs/source/fastNLP.modules.encoder.transformer.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.transformer -=================================== - -.. automodule:: fastNLP.modules.encoder.transformer - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.encoder.variational_rnn.rst b/docs/source/fastNLP.modules.encoder.variational_rnn.rst deleted file mode 100644 index 348fb3d8..00000000 --- a/docs/source/fastNLP.modules.encoder.variational_rnn.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.modules.encoder.variational\_rnn -======================================== - -.. automodule:: fastNLP.modules.encoder.variational_rnn - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.modules.rst b/docs/source/fastNLP.modules.rst index d04ccdcf..646ef2d3 100644 --- a/docs/source/fastNLP.modules.rst +++ b/docs/source/fastNLP.modules.rst @@ -2,16 +2,16 @@ fastNLP.modules =============== .. automodule:: fastNLP.modules - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: 子模块 ----------- .. toctree:: - :titlesonly: + :titlesonly: + :maxdepth: 1 - fastNLP.modules.aggregator - fastNLP.modules.decoder - fastNLP.modules.encoder \ No newline at end of file + fastNLP.modules.decoder + fastNLP.modules.encoder \ No newline at end of file diff --git a/docs/source/fastNLP.rst b/docs/source/fastNLP.rst index f0c3d41c..0057a184 100644 --- a/docs/source/fastNLP.rst +++ b/docs/source/fastNLP.rst @@ -2,19 +2,18 @@ API 文档 =============== .. automodule:: fastNLP - :members: - :undoc-members: - :show-inheritance: + :members: + :undoc-members: + :show-inheritance: 内部模块 ----------- .. toctree:: - :titlesonly: - :maxdepth: 3 - - fastNLP.core - fastNLP.io - fastNLP.modules - fastNLP.models + :maxdepth: 1 + fastNLP.core + fastNLP.embeddings + fastNLP.io + fastNLP.models + fastNLP.modules diff --git a/docs/source/figures/text_classification.png b/docs/source/figures/text_classification.png index 0d36a2a1..21502708 100644 Binary files a/docs/source/figures/text_classification.png and b/docs/source/figures/text_classification.png differ diff --git a/docs/source/figures/workflow.png b/docs/source/figures/workflow.png index d2f22df8..d8e4e455 100644 Binary files a/docs/source/figures/workflow.png and b/docs/source/figures/workflow.png differ diff --git a/docs/source/index.rst b/docs/source/index.rst index da510437..d48af986 100644 --- a/docs/source/index.rst +++ b/docs/source/index.rst @@ -1,62 +1,28 @@ fastNLP 中文文档 ===================== -fastNLP 是一款轻量级的 NLP 处理套件。你既可以使用它快速地完成一个命名实体识别（NER）、中文分词或文本分类任务； -也可以使用他构建许多复杂的网络模型，进行科研。它具有如下的特性: +`fastNLP `_ 是一款轻量级的 NLP 处理套件。你既可以使用它快速地完成一个序列标注 +（NER、POS-Tagging等）、中文分词、文本分类、Matching、指代消解、摘要等任务 +（详见 `reproduction `_ ）； +也可以使用它构建许多复杂的网络模型，进行科研。它具有如下的特性： -- 统一的Tabular式数据容器，让数据预处理过程简洁明了。内置多种数据集的DataSet Loader，省去预处理代码。 -- 各种方便的NLP工具，例如预处理embedding加载; 中间数据cache等; -- 详尽的中文文档以供查阅； -- 提供诸多高级模块，例如Variational LSTM, Transformer, CRF等; -- 封装CNNText，Biaffine等模型可供直接使用; -- 便捷且具有扩展性的训练器; 提供多种内置callback函数，方便实验记录、异常捕获等。 +- 统一的Tabular式数据容器，让数据预处理过程简洁明了。内置多种数据集的 :mod:`~fastNLP.io.data_loader` ，省去预处理代码; +- 多种训练、测试组件，例如训练器 :class:`~fastNLP.Trainer` ；测试器 :class:`~fastNLP.Tester` ；以及各种评测 :mod:`~fastNLP.core.metrics` 等等; +- 各种方便的NLP工具，例如预处理 :mod:`embedding` 加载（包括ELMo和BERT）; 中间数据存储 :func:`cache ` 等; +- 提供诸多高级模块 :mod:`~fastNLP.modules`，例如 :class:`~fastNLP.modules.VarLSTM` , :class:`Transformer` , :class:`CRF` 等; +- 在序列标注、中文分词、文本分类、Matching、指代消解、摘要等任务上封装了各种 :mod:`~fastNLP.models` 可供直接使用; +- 训练器便捷且具有扩展性，提供多种内置 :mod:`~fastNLP.core.callback` 函数，方便实验记录、异常捕获等。 -内置组件 ------------- - -大部分用于的 NLP 任务神经网络都可以看做由编码（encoder）、聚合（aggregator）、解码（decoder）三种模块组成。 - -.. image:: figures/text_classification.png - -fastNLP 在 :mod:`~fastNLP.modules` 模块中内置了三种模块的诸多组件，可以帮助用户快速搭建自己所需的网络。 -三种模块的功能和常见组件如下: - -+-----------------------+-----------------------+-----------------------+ -| module type | functionality | example | -+=======================+=======================+=======================+ -| encoder | 将输入编码为具有具 | embedding, RNN, CNN, | -| | 有表示能力的向量 | transformer | -+-----------------------+-----------------------+-----------------------+ -| aggregator | 从多个向量中聚合信息 | self-attention, | -| | | max-pooling | -+-----------------------+-----------------------+-----------------------+ -| decoder | 将具有某种表示意义的 | MLP, CRF | -| | 向量解码为需要的输出 | | -| | 形式 | | -+-----------------------+-----------------------+-----------------------+ - - -内置模型 ----------------- - -fastNLP 在 :mod:`~fastNLP.models` 模块中内置了如 :class:`~fastNLP.models.CNNText` 、 -:class:`~fastNLP.models.SeqLabeling` 等完整的模型，以供用户直接使用。 - -.. todo:: - 这些模型的介绍如下表所示：（模型名称 + 介绍 + 任务上的结果） - 用户手册 ---------------- .. toctree:: - :maxdepth: 1 + :maxdepth: 2 - 安装指南 - 快速入门 - 详细指南 - 科研指南 - 注释语法 + 安装指南 + 快速入门 + 详细教程 API 文档 ------------- @@ -69,11 +35,11 @@ API 文档 fastNLP -fitlog ------- +fitlog文档 +---------- -用户可以 `点此 `_ 查看fitlog的文档。 -fitlog 是由我们团队开发，用于帮助用户记录日志并管理代码的工具 +您可以 `点此 `_ 查看fitlog的文档。 +fitlog 是由我们团队开发的日志记录+代码管理的工具。 索引与搜索 ================== diff --git a/docs/source/user/with_fitlog.rst b/docs/source/tutorials/tutorial_10_fitlog.rst similarity index 96% rename from docs/source/user/with_fitlog.rst rename to docs/source/tutorials/tutorial_10_fitlog.rst index 51445775..0fa24143 100644 --- a/docs/source/user/with_fitlog.rst +++ b/docs/source/tutorials/tutorial_10_fitlog.rst @@ -1,6 +1,6 @@ -================= -科研向导 -================= +============================================ +使用fitlog 辅助 fastNLP 进行科研 +============================================ 本文介绍结合使用 fastNLP 和 fitlog 进行科研的方法。 diff --git a/docs/source/tutorials/tutorial_1_data_preprocess.rst b/docs/source/tutorials/tutorial_1_data_preprocess.rst new file mode 100644 index 00000000..0ec63f87 --- /dev/null +++ b/docs/source/tutorials/tutorial_1_data_preprocess.rst @@ -0,0 +1,156 @@ +============================== +使用DataSet预处理文本 +============================== + +:class:`~fastNLP.DataSet` 是fastNLP中用于承载数据的容器。可以将DataSet看做是一个表格， +每一行是一个sample (在fastNLP中被称为 :mod:`~fastNLP.core.instance` )， +每一列是一个feature (在fastNLP中称为 :mod:`~fastNLP.core.field` )。 + +.. csv-table:: + :header: "sentence", "words", "seq_len" + + "This is the first instance .", "[This, is, the, first, instance, .]", 6 + "Second instance .", "[Second, instance, .]", 3 + "Third instance .", "[Third, instance, .]", 3 + "...", "[...]", "..." + +上面是一个样例数据中 DataSet 的存储结构。其中它的每一行是一个 :class:`~fastNLP.Instance` 对象； 每一列是一个 :class:`~fastNLP.FieldArray` 对象。 + + +----------------------------- +数据集构建和删除 +----------------------------- + +我们使用传入字典的方式构建一个数据集，这是 :class:`~fastNLP.DataSet` 初始化的最基础的方式 + +.. code-block:: python + + from fastNLP import DataSet + data = {'sentence':["This is the first instance .", "Second instance .", "Third instance ."], + 'words': [['this', 'is', 'the', 'first', 'instance', '.'], ['Second', 'instance', '.'], ['Third', 'instance', '.']], + 'seq_len': [6, 3, 3]} + dataset = DataSet(data) + # 传入的dict的每个key的value应该为具有相同长度的list + +我们还可以使用 :func:`~fastNLP.DataSet.append` 方法向数据集内增加数据 + +.. code-block:: python + + from fastNLP import DataSet + from fastNLP import Instance + dataset = DataSet() + instance = Instance(sentence="This is the first instance", + words=['this', 'is', 'the', 'first', 'instance', '.'], + seq_len=6) + dataset.append(instance) + # 可以继续append更多内容，但是append的instance应该和前面的instance拥有完全相同的field + +另外，我们还可以用 :class:`~fastNLP.Instance` 数组的方式构建数据集 + +.. code-block:: python + + from fastNLP import DataSet + from fastNLP import Instance + dataset = DataSet([ + Instance(sentence="This is the first instance", + words=['this', 'is', 'the', 'first', 'instance', '.'], + seq_len=6), + Instance(sentence="Second instance .", + words=['Second', 'instance', '.'], + seq_len=3) + ]) + +在初步构建完数据集之后，我们可以通过 `for` 循环遍历 :class:`~fastNLP.DataSet` 中的内容。 + +.. code-block:: python + + for instance in dataset: + # do something + +FastNLP 同样提供了多种删除数据的方法 :func:`~fastNLP.DataSet.drop` 、 :func:`~fastNLP.DataSet.delete_instance` 和 :func:`~fastNLP.DataSet.delete_field` + +.. code-block:: python + + from fastNLP import DataSet + dataset = DataSet({'a': list(range(-5, 5))}) + # 返回满足条件的instance,并放入DataSet中 + dropped_dataset = dataset.drop(lambda ins:ins['a']<0, inplace=False) + # 在dataset中删除满足条件的instance + dataset.drop(lambda ins:ins['a']<0) # dataset的instance数量减少 + # 删除第3个instance + dataset.delete_instance(2) + # 删除名为'a'的field + dataset.delete_field('a') + +----------------------------- +简单的数据预处理 +----------------------------- + +因为 fastNLP 中的数据是按列存储的，所以大部分的数据预处理操作是以列（ :mod:`~fastNLP.core.field` ）为操作对象的。 +首先，我们可以检查特定名称的 :mod:`~fastNLP.core.field` 是否存在，并对其进行改名。 + +.. code-block:: python + + # 检查是否存在名为'a'的field + dataset.has_field('a') # 或 ('a' in dataset) + # 将名为'a'的field改名为'b' + dataset.rename_field('a', 'b') + # DataSet的长度 + len(dataset) + +其次，我们可以使用 :func:`~fastNLP.DataSet.apply` 或 :func:`~fastNLP.DataSet.apply_field` 进行数据预处理操作操作。 +这两个方法通过传入一个对单一 :mod:`~fastNLP.core.instance` 操作的函数， +自动地帮助你对一个 :mod:`~fastNLP.core.field` 中的每个 :mod:`~fastNLP.core.instance` 调用这个函数，完成整体的操作。 +这个传入的函数可以是 lambda 匿名函数，也可以是完整定义的函数。同时，你还可以用 ``new_field_name`` 参数指定数据处理后存储的 :mod:`~fastNLP.core.field` 的名称。 + +.. code-block:: python + + from fastNLP import DataSet + data = {'sentence':["This is the first instance .", "Second instance .", "Third instance ."]} + dataset = DataSet(data) + + # 将句子分成单词形式, 详见DataSet.apply()方法 + dataset.apply(lambda ins: ins['sentence'].split(), new_field_name='words') + + # 或使用DataSet.apply_field() + dataset.apply_field(lambda sent:sent.split(), field_name='sentence', new_field_name='words') + + # 除了匿名函数，也可以定义函数传递进去 + def get_words(instance): + sentence = instance['sentence'] + words = sentence.split() + return words + dataset.apply(get_words, new_field_name='words') + +除了手动处理数据集之外，你还可以使用 fastNLP 提供的各种 :class:`~fastNLP.io.base_loader.DataSetLoader` 来进行数据处理。 +详细请参考这篇教程 :doc:`使用DataSetLoader加载数据集 ` 。 + +----------------------------- +DataSet与pad +----------------------------- + +在fastNLP里，pad是与一个 :mod:`~fastNLP.core.field` 绑定的。即不同的 :mod:`~fastNLP.core.field` 可以使用不同的pad方式，比如在英文任务中word需要的pad和 +character的pad方式往往是不同的。fastNLP是通过一个叫做 :class:`~fastNLP.Padder` 的子类来完成的。 +默认情况下，所有field使用 :class:`~fastNLP.AutoPadder` +。可以通过使用以下方式设置Padder(如果将padder设置为None，则该field不会进行pad操作)。 +大多数情况下直接使用 :class:`~fastNLP.AutoPadder` 就可以了。 +如果 :class:`~fastNLP.AutoPadder` 或 :class:`~fastNLP.EngChar2DPadder` 无法满足需求， +也可以自己写一个 :class:`~fastNLP.Padder` 。 + +.. code-block:: python + + from fastNLP import DataSet + from fastNLP import EngChar2DPadder + import random + dataset = DataSet() + max_chars, max_words, sent_num = 5, 10, 20 + contents = [[ + [random.randint(1, 27) for _ in range(random.randint(1, max_chars))] + for _ in range(random.randint(1, max_words)) + ] for _ in range(sent_num)] + # 初始化时传入 + dataset.add_field('chars', contents, padder=EngChar2DPadder()) + # 直接设置 + dataset.set_padder('chars', EngChar2DPadder()) + # 也可以设置pad的value + dataset.set_pad_val('chars', -1) diff --git a/docs/source/tutorials/tutorial_2_load_dataset.rst b/docs/source/tutorials/tutorial_2_load_dataset.rst new file mode 100644 index 00000000..4fa4a84d --- /dev/null +++ b/docs/source/tutorials/tutorial_2_load_dataset.rst @@ -0,0 +1,224 @@ +================================= +使用DataSetLoader加载数据集 +================================= + +这一部分是一个关于如何加载数据集的教程 + +教程目录： + + - `Part I: 数据集容器`_ + - `Part II: 数据集的使用方式`_ + - `Part III: 不同数据类型的DataSetLoader`_ + - `Part IV: DataSetLoader举例`_ + - `Part V: fastNLP封装好的数据集加载器`_ + + +---------------------------- +Part I: 数据集容器 +---------------------------- + +在fastNLP中，我们使用 :class:`~fastNLP.io.base_loader.DataBundle` 来存储数据集信息。 +:class:`~fastNLP.io.base_loader.DataBundle` 类包含了两个重要内容： `datasets` 和 `vocabs` 。 + +`datasets` 是一个 `key` 为数据集名称（如 `train` ， `dev` ，和 `test` 等）， `value` 为 :class:`~fastNLP.DataSet` 的字典。 + +`vocabs` 是一个 `key` 为词表名称（如 :attr:`fastNLP.Const.INPUT` 表示输入文本的词表名称， :attr:`fastNLP.Const.TARGET` 表示目标 +的真实标签词表的名称，等等）， `value` 为词表内容（ :class:`~fastNLP.Vocabulary` ）的字典。 + +---------------------------- +Part II: 数据集的使用方式 +---------------------------- + +在fastNLP中，我们采用 :class:`~fastNLP.io.base_loader.DataSetLoader` 来作为加载数据集的基类。 +:class:`~fastNLP.io.base_loader.DataSetLoader` 定义了各种DataSetLoader所需的API接口，开发者应该继承它实现各种的DataSetLoader。 +在各种数据集的DataSetLoader当中，至少应该编写如下内容: + + - _load 函数：从一个数据文件中读取数据到一个 :class:`~fastNLP.DataSet` + - load 函数（可以使用基类的方法）：从一个或多个数据文件中读取数据到一个或多个 :class:`~fastNLP.DataSet` + - process 函数：一个或多个从数据文件中读取数据，并处理成可以训练的 :class:`~fastNLP.io.DataBundle` + + **\*process函数中可以调用load函数或_load函数** + +DataSetLoader的_load或者load函数返回的 :class:`~fastNLP.DataSet` 当中，内容为数据集的文本信息，process函数返回的 +:class:`~fastNLP.io.DataBundle` 当中， `datasets` 的内容为已经index好的、可以直接被 :class:`~fastNLP.Trainer` +接受的内容。 + +-------------------------------------------------------- +Part III: 不同数据类型的DataSetLoader +-------------------------------------------------------- + +:class:`~fastNLP.io.dataset_loader.CSVLoader` + 读取CSV类型的数据集文件。例子如下： + + .. code-block:: python + + data_set_loader = CSVLoader( + headers=('words', 'target'), sep='\t' + ) + # 表示将CSV文件中每一行的第一项填入'words' field，第二项填入'target' field。 + # 其中每两项之间由'\t'分割开来 + + data_set = data_set_loader._load('path/to/your/file') + + 数据集内容样例如下 :: + + But it does not leave you with much . 1 + You could hate it for the same reason . 1 + The performances are an absolute joy . 4 + + +:class:`~fastNLP.io.dataset_loader.JsonLoader` + 读取Json类型的数据集文件，数据必须按行存储，每行是一个包含各类属性的Json对象。例子如下： + + .. code-block:: python + + data_set_loader = JsonLoader( + fields={'sentence1': 'words1', 'sentence2': 'words2', 'gold_label': 'target'} + ) + # 表示将Json对象中'sentence1'、'sentence2'和'gold_label'对应的值赋给'words1'、'words2'、'target'这三个fields + + data_set = data_set_loader._load('path/to/your/file') + + 数据集内容样例如下 :: + + {"annotator_labels": ["neutral"], "captionID": "3416050480.jpg#4", "gold_label": "neutral", "pairID": "3416050480.jpg#4r1n", "sentence1": "A person on a horse jumps over a broken down airplane.", "sentence1_binary_parse": "( ( ( A person ) ( on ( a horse ) ) ) ( ( jumps ( over ( a ( broken ( down airplane ) ) ) ) ) . ) )", "sentence1_parse": "(ROOT (S (NP (NP (DT A) (NN person)) (PP (IN on) (NP (DT a) (NN horse)))) (VP (VBZ jumps) (PP (IN over) (NP (DT a) (JJ broken) (JJ down) (NN airplane)))) (. .)))", "sentence2": "A person is training his horse for a competition.", "sentence2_binary_parse": "( ( A person ) ( ( is ( ( training ( his horse ) ) ( for ( a competition ) ) ) ) . ) )", "sentence2_parse": "(ROOT (S (NP (DT A) (NN person)) (VP (VBZ is) (VP (VBG training) (NP (PRP$ his) (NN horse)) (PP (IN for) (NP (DT a) (NN competition))))) (. .)))"} + {"annotator_labels": ["contradiction"], "captionID": "3416050480.jpg#4", "gold_label": "contradiction", "pairID": "3416050480.jpg#4r1c", "sentence1": "A person on a horse jumps over a broken down airplane.", "sentence1_binary_parse": "( ( ( A person ) ( on ( a horse ) ) ) ( ( jumps ( over ( a ( broken ( down airplane ) ) ) ) ) . ) )", "sentence1_parse": "(ROOT (S (NP (NP (DT A) (NN person)) (PP (IN on) (NP (DT a) (NN horse)))) (VP (VBZ jumps) (PP (IN over) (NP (DT a) (JJ broken) (JJ down) (NN airplane)))) (. .)))", "sentence2": "A person is at a diner, ordering an omelette.", "sentence2_binary_parse": "( ( A person ) ( ( ( ( is ( at ( a diner ) ) ) , ) ( ordering ( an omelette ) ) ) . ) )", "sentence2_parse": "(ROOT (S (NP (DT A) (NN person)) (VP (VBZ is) (PP (IN at) (NP (DT a) (NN diner))) (, ,) (S (VP (VBG ordering) (NP (DT an) (NN omelette))))) (. .)))"} + {"annotator_labels": ["entailment"], "captionID": "3416050480.jpg#4", "gold_label": "entailment", "pairID": "3416050480.jpg#4r1e", "sentence1": "A person on a horse jumps over a broken down airplane.", "sentence1_binary_parse": "( ( ( A person ) ( on ( a horse ) ) ) ( ( jumps ( over ( a ( broken ( down airplane ) ) ) ) ) . ) )", "sentence1_parse": "(ROOT (S (NP (NP (DT A) (NN person)) (PP (IN on) (NP (DT a) (NN horse)))) (VP (VBZ jumps) (PP (IN over) (NP (DT a) (JJ broken) (JJ down) (NN airplane)))) (. .)))", "sentence2": "A person is outdoors, on a horse.", "sentence2_binary_parse": "( ( A person ) ( ( ( ( is outdoors ) , ) ( on ( a horse ) ) ) . ) )", "sentence2_parse": "(ROOT (S (NP (DT A) (NN person)) (VP (VBZ is) (ADVP (RB outdoors)) (, ,) (PP (IN on) (NP (DT a) (NN horse)))) (. .)))"} + +------------------------------------------ +Part IV: DataSetLoader举例 +------------------------------------------ + +以Matching任务为例子： + + :class:`~fastNLP.io.data_loader.MatchingLoader` + 我们在fastNLP当中封装了一个Matching任务数据集的数据加载类： :class:`~fastNLP.io.data_loader.MatchingLoader` . + + 在MatchingLoader类当中我们封装了一个对数据集中的文本内容进行进一步的预处理的函数： + :meth:`~fastNLP.io.data_loader.MatchingLoader.process` + 这个函数具有各种预处理option，如： + - 是否将文本转成全小写 + - 是否需要序列长度信息，需要什么类型的序列长度信息 + - 是否需要用BertTokenizer来获取序列的WordPiece信息 + - 等等 + + 具体内容参见 :meth:`fastNLP.io.MatchingLoader.process` 。 + + :class:`~fastNLP.io.data_loader.SNLILoader` + 一个关于SNLI数据集的DataSetLoader。SNLI数据集来自 + `SNLI Data Set `_ . + + 在 :class:`~fastNLP.io.data_loader.SNLILoader` 的 :meth:`~fastNLP.io.data_loader.SNLILoader._load` + 函数中，我们用以下代码将数据集内容从文本文件读入内存： + + .. code-block:: python + + data = SNLILoader().process( + paths='path/to/snli/data', to_lower=False, seq_len_type='seq_len', + get_index=True, concat=False, + ) + print(data) + + 输出的内容是:: + + In total 3 datasets: + train has 549367 instances. + dev has 9842 instances. + test has 9824 instances. + In total 2 vocabs: + words has 43154 entries. + target has 3 entries. + + + 这里的data是一个 :class:`~fastNLP.io.base_loader.DataBundle` ，取 ``datasets`` 字典里的内容即可直接传入 + :class:`~fastNLP.Trainer` 或者 :class:`~fastNLP.Tester` 进行训练或者测试。 + + :class:`~fastNLP.io.data_loader.IMDBLoader` + 以IMDB数据集为例，在 :class:`~fastNLP.io.data_loader.IMDBLoader` 的 :meth:`~fastNLP.io.data_loader.IMDBLoader._load` + 函数中，我们用以下代码将数据集内容从文本文件读入内存： + + .. code-block:: python + + data = IMDBLoader().process( + paths={'train': 'path/to/train/file', 'test': 'path/to/test/file'} + ) + print(data) + + 输出的内容是:: + + In total 3 datasets: + train has 22500 instances. + test has 25000 instances. + dev has 2500 instances. + In total 2 vocabs: + words has 82846 entries. + target has 2 entries. + + + 这里的将原来的train集按9:1的比例分成了训练集和验证集。 + + +------------------------------------------ +Part V: fastNLP封装好的数据集加载器 +------------------------------------------ + +fastNLP封装好的数据集加载器可以适用于多种类型的任务： + + - `文本分类任务`_ + - `序列标注任务`_ + - `Matching任务`_ + + +文本分类任务 +------------------- + +========================== ================================================================== +数据集名称 数据集加载器 +-------------------------- ------------------------------------------------------------------ +IMDb :class:`~fastNLP.io.data_loader.IMDBLoader` +-------------------------- ------------------------------------------------------------------ +SST :class:`~fastNLP.io.data_loader.SSTLoader` +-------------------------- ------------------------------------------------------------------ +SST-2 :class:`~fastNLP.io.data_loader.SST2Loader` +-------------------------- ------------------------------------------------------------------ +Yelp Polarity :class:`~fastNLP.io.data_loader.YelpLoader` +-------------------------- ------------------------------------------------------------------ +Yelp Full :class:`~fastNLP.io.data_loader.YelpLoader` +-------------------------- ------------------------------------------------------------------ +MTL16 :class:`~fastNLP.io.data_loader.MTL16Loader` +========================== ================================================================== + + + +序列标注任务 +------------------- + +========================== ================================================================== +数据集名称 数据集加载器 +-------------------------- ------------------------------------------------------------------ +Conll :class:`~fastNLP.io.data_loader.ConllLoader` +-------------------------- ------------------------------------------------------------------ +Conll2003 :class:`~fastNLP.io.data_loader.Conll2003Loader` +-------------------------- ------------------------------------------------------------------ +人民日报数据集 :class:`~fastNLP.io.data_loader.PeopleDailyCorpusLoader` +========================== ================================================================== + + + +Matching任务 +------------------- + +========================== ================================================================== +数据集名称 数据集加载器 +-------------------------- ------------------------------------------------------------------ +SNLI :class:`~fastNLP.io.data_loader.SNLILoader` +-------------------------- ------------------------------------------------------------------ +MultiNLI :class:`~fastNLP.io.data_loader.MNLILoader` +-------------------------- ------------------------------------------------------------------ +QNLI :class:`~fastNLP.io.data_loader.QNLILoader` +-------------------------- ------------------------------------------------------------------ +RTE :class:`~fastNLP.io.data_loader.RTELoader` +-------------------------- ------------------------------------------------------------------ +Quora Pair Dataset :class:`~fastNLP.io.data_loader.QuoraLoader` +========================== ================================================================== + diff --git a/docs/source/tutorials/tutorial_3_embedding.rst b/docs/source/tutorials/tutorial_3_embedding.rst new file mode 100644 index 00000000..489b43b4 --- /dev/null +++ b/docs/source/tutorials/tutorial_3_embedding.rst @@ -0,0 +1,214 @@ +========================================= +使用Embedding模块将文本转成向量 +========================================= + +这一部分是一个关于在fastNLP当中使用embedding的教程。 + +教程目录： + + - `Part I: embedding介绍`_ + - `Part II: 使用随机初始化的embedding`_ + - `Part III: 使用预训练的静态embedding`_ + - `Part IV: 使用预训练的Contextual Embedding(ELMo & BERT)`_ + - `Part V: 使用character-level的embedding`_ + - `Part VI: 叠加使用多个embedding`_ + + + + +--------------------------------------- +Part I: embedding介绍 +--------------------------------------- + +与torch.nn.Embedding类似，fastNLP的embedding接受的输入是一个被index好的序列，输出的内容是这个序列的embedding结果。 + +fastNLP的embedding包括了预训练embedding和随机初始化embedding。 + + +--------------------------------------- +Part II: 使用随机初始化的embedding +--------------------------------------- + +使用随机初始化的embedding参见 :class:`~fastNLP.embeddings.embedding.Embedding` 。 + +可以传入词表大小和embedding维度： + +.. code-block:: python + + embed = Embedding(10000, 50) + +也可以传入一个初始化的参数矩阵： + +.. code-block:: python + + embed = Embedding(init_embed) + +其中的init_embed可以是torch.FloatTensor、torch.nn.Embedding或者numpy.ndarray。 + + +--------------------------------------- +Part III: 使用预训练的静态embedding +--------------------------------------- + +在使用预训练的embedding之前，需要根据数据集的内容构建一个词表 :class:`~fastNLP.core.vocabulary.Vocabulary` ，在 +预训练embedding类初始化的时候需要将这个词表作为参数传入。 + +在fastNLP中，我们提供了 :class:`~fastNLP.embeddings.StaticEmbedding` 这一个类。 +通过 :class:`~fastNLP.embeddings.StaticEmbedding` 可以加载预训练好的静态 +Embedding，例子如下： + +.. code-block:: python + + embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) + +vocab为根据数据集构建的词表，model_dir_or_name可以是一个路径，也可以是embedding模型的名称： + + 1 如果传入的是路径，那么fastNLP将会根据该路径来读取预训练的权重文件并将embedding加载进来(glove + 和word2vec类型的权重文件都支持) + + 2 如果传入的是模型名称，那么fastNLP将会根据名称查找embedding模型，如果在cache目录下找到模型则会 + 自动加载；如果找不到则会自动下载。可以通过环境变量 ``FASTNLP_CACHE_DIR`` 来自定义cache目录，如:: + + $ FASTNLP_CACHE_DIR=~/fastnlp_cache_dir python your_python_file.py + +这个命令表示fastNLP将会在 `~/fastnlp_cache_dir` 这个目录下寻找模型，找不到则会自动将模型下载到这个目录 + +目前支持的静态embedding模型有： + + ========================== ================================ + 模型名称 模型 + -------------------------- -------------------------------- + en glove.840B.300d + -------------------------- -------------------------------- + en-glove-840d-300 glove.840B.300d + -------------------------- -------------------------------- + en-glove-6b-50 glove.6B.50d + -------------------------- -------------------------------- + en-word2vec-300 谷歌word2vec 300维 + -------------------------- -------------------------------- + en-fasttext 英文fasttext 300维 + -------------------------- -------------------------------- + cn 腾讯中文词向量 200维 + -------------------------- -------------------------------- + cn-fasttext 中文fasttext 300维 + ========================== ================================ + + + +----------------------------------------------------------- +Part IV: 使用预训练的Contextual Embedding(ELMo & BERT) +----------------------------------------------------------- + +在fastNLP中，我们提供了ELMo和BERT的embedding： :class:`~fastNLP.embeddings.ElmoEmbedding` +和 :class:`~fastNLP.embeddings.BertEmbedding` 。 + +与静态embedding类似，ELMo的使用方法如下： + +.. code-block:: python + + embed = ElmoEmbedding(vocab, model_dir_or_name='small', requires_grad=False) + +目前支持的ElmoEmbedding模型有： + + ========================== ================================ + 模型名称 模型 + -------------------------- -------------------------------- + small allennlp ELMo的small + -------------------------- -------------------------------- + medium allennlp ELMo的medium + -------------------------- -------------------------------- + original allennlp ELMo的original + -------------------------- -------------------------------- + 5.5b-original allennlp ELMo的5.5B original + ========================== ================================ + +BERT-embedding的使用方法如下： + +.. code-block:: python + + embed = BertEmbedding( + vocab, model_dir_or_name='en-base-cased', requires_grad=False, layers='4,-2,-1' + ) + +其中layers变量表示需要取哪几层的encode结果。 + +目前支持的BertEmbedding模型有： + + ========================== ==================================== + 模型名称 模型 + -------------------------- ------------------------------------ + en bert-base-cased + -------------------------- ------------------------------------ + en-base-uncased bert-base-uncased + -------------------------- ------------------------------------ + en-base-cased bert-base-cased + -------------------------- ------------------------------------ + en-large-uncased bert-large-uncased + -------------------------- ------------------------------------ + en-large-cased bert-large-cased + -------------------------- ------------------------------------ + -------------------------- ------------------------------------ + en-large-cased-wwm bert-large-cased-whole-word-mask + -------------------------- ------------------------------------ + en-large-uncased-wwm bert-large-uncased-whole-word-mask + -------------------------- ------------------------------------ + en-base-cased-mrpc bert-base-cased-finetuned-mrpc + -------------------------- ------------------------------------ + -------------------------- ------------------------------------ + multilingual bert-base-multilingual-cased + -------------------------- ------------------------------------ + multilingual-base-uncased bert-base-multilingual-uncased + -------------------------- ------------------------------------ + multilingual-base-cased bert-base-multilingual-cased + ========================== ==================================== + +----------------------------------------------------- +Part V: 使用character-level的embedding +----------------------------------------------------- + +除了预训练的embedding以外，fastNLP还提供了CharEmbedding： :class:`~fastNLP.embeddings.CNNCharEmbedding` 和 +:class:`~fastNLP.embeddings.LSTMCharEmbedding` 。 + +CNNCharEmbedding的使用例子如下： + +.. code-block:: python + + embed = CNNCharEmbedding(vocab, embed_size=100, char_emb_size=50) + +这表示这个CNNCharEmbedding当中character的embedding维度大小为50，返回的embedding结果维度大小为100。 + +与CNNCharEmbedding类似，LSTMCharEmbedding的使用例子如下： + +.. code-block:: python + + embed = LSTMCharEmbedding(vocab, embed_size=100, char_emb_size=50) + +这表示这个LSTMCharEmbedding当中character的embedding维度大小为50，返回的embedding结果维度大小为100。 + + + +----------------------------------------------------- +Part VI: 叠加使用多个embedding +----------------------------------------------------- + +在fastNLP中，我们使用 :class:`~fastNLP.embeddings.StackEmbedding` 来叠加多个embedding + +例子如下： + +.. code-block:: python + + embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) + embed_2 = StaticEmbedding(vocab, model_dir_or_name='en-word2vec-300', requires_grad=True) + + stack_embed = StackEmbedding([embed_1, embed_2]) + +StackEmbedding会把多个embedding的结果拼接起来，如上面例子的stack_embed返回的embedding维度为350维。 + +除此以外，还可以把静态embedding跟上下文相关的embedding拼接起来： + +.. code-block:: python + + elmo_embedding = ElmoEmbedding(vocab, model_dir_or_name='medium', layers='0,1,2', requires_grad=False) + glove_embedding = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) + + stack_embed = StackEmbedding([elmo_embedding, glove_embedding]) diff --git a/docs/source/tutorials/tutorial_4_loss_optimizer.rst b/docs/source/tutorials/tutorial_4_loss_optimizer.rst new file mode 100644 index 00000000..a6e1730a --- /dev/null +++ b/docs/source/tutorials/tutorial_4_loss_optimizer.rst @@ -0,0 +1,267 @@ +============================================================================== +动手实现一个文本分类器I-使用Trainer和Tester快速训练和测试 +============================================================================== + +我们使用和 :doc:`/user/quickstart` 中一样的任务来进行详细的介绍。给出一段评价性文字，预测其情感倾向是积极（label=1）、 +消极（label=0）还是中性（label=2），使用 :class:`~fastNLP.Trainer` 和 :class:`~fastNLP.Tester` 来进行快速训练和测试。 + +-------------- +数据处理 +-------------- + +数据读入 + 我们可以使用 fastNLP :mod:`fastNLP.io` 模块中的 :class:`~fastNLP.io.SSTLoader` 类，轻松地读取SST数据集（数据来源：https://nlp.stanford.edu/sentiment/trainDevTestTrees_PTB.zip）。 + 这里的 dataset 是 fastNLP 中 :class:`~fastNLP.DataSet` 类的对象。 + + .. code-block:: python + + from fastNLP.io import SSTLoader + + loader = SSTLoader() + #这里的all.txt是下载好数据后train.txt、dev.txt、test.txt的组合 + dataset = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + print(dataset[0]) + + 输出数据如下:: + + {'words': ['It', "'s", 'a', 'lovely', 'film', 'with', 'lovely', 'performances', 'by', 'Buy', 'and', 'Accorsi', '.'] type=list, + 'target': positive type=str} + + 除了读取数据外，fastNLP 还提供了读取其它文件类型的 Loader 类、读取 Embedding的 Loader 等。详见 :doc:`/fastNLP.io` 。 + + +数据处理 + 我们使用 :class:`~fastNLP.DataSet` 类的 :meth:`~fastNLP.DataSet.apply` 方法将 ``target`` :mod:`~fastNLP.core.field` 转化为整数。 + + .. code-block:: python + + def label_to_int(x): + if x['target']=="positive": + return 1 + elif x['target']=="negative": + return 0 + else: + return 2 + + # 将label转为整数 + dataset.apply(lambda x: label_to_int(x), new_field_name='target') + + ``words`` 和 ``target`` 已经足够用于 :class:`~fastNLP.models.CNNText` 的训练了，但我们从其文档 + :class:`~fastNLP.models.CNNText` 中看到，在 :meth:`~fastNLP.models.CNNText.forward` 的时候，还可以传入可选参数 ``seq_len`` 。 + 所以，我们再使用 :meth:`~fastNLP.DataSet.apply_field` 方法增加一个名为 ``seq_len`` 的 :mod:`~fastNLP.core.field` 。 + + .. code-block:: python + + # 增加长度信息 + dataset.apply_field(lambda x: len(x), field_name='words', new_field_name='seq_len') + + 观察可知： :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 类似， + 但所传入的 `lambda` 函数是针对一个 :class:`~fastNLP.Instance` 中的一个 :mod:`~fastNLP.core.field` 的； + 而 :meth:`~fastNLP.DataSet.apply` 所传入的 `lambda` 函数是针对整个 :class:`~fastNLP.Instance` 的。 + + .. note:: + `lambda` 函数即匿名函数，是 Python 的重要特性。 ``lambda x: len(x)`` 和下面的这个函数的作用相同:: + + def func_lambda(x): + return len(x) + + 你也可以编写复杂的函数做为 :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 的参数 + +Vocabulary 的使用 + 我们再用 :class:`~fastNLP.Vocabulary` 类来统计数据中出现的单词，并使用 :meth:`~fastNLP.Vocabulary.index_dataset` + 将单词序列转化为训练可用的数字序列。 + + .. code-block:: python + + from fastNLP import Vocabulary + + # 使用Vocabulary类统计单词，并将单词序列转化为数字序列 + vocab = Vocabulary(min_freq=2).from_dataset(dataset, field_name='words') + vocab.index_dataset(dataset, field_name='words',new_field_name='words') + print(dataset[0]) + + 输出数据如下:: + + {'words': [27, 9, 6, 913, 16, 18, 913, 124, 31, 5715, 5, 1, 2] type=list, + 'target': 1 type=int, + 'seq_len': 13 type=int} + + +--------------------- +使用内置模型训练 +--------------------- + +内置模型的输入输出命名 + fastNLP内置了一些完整的神经网络模型，详见 :doc:`/fastNLP.models` , 我们使用其中的 :class:`~fastNLP.models.CNNText` 模型进行训练。 + 为了使用内置的 :class:`~fastNLP.models.CNNText`，我们必须修改 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 的名称。 + 在这个例子中模型输入 (forward方法的参数) 为 ``words`` 和 ``seq_len`` ; 预测输出为 ``pred`` ;标准答案为 ``target`` 。 + 具体的命名规范可以参考 :doc:`/fastNLP.core.const` 。 + + 如果不想查看文档，您也可以使用 :class:`~fastNLP.Const` 类进行命名。下面的代码展示了给 :class:`~fastNLP.DataSet` 中 + :mod:`~fastNLP.core.field` 改名的 :meth:`~fastNLP.DataSet.rename_field` 方法，以及 :class:`~fastNLP.Const` 类的使用方法。 + + .. code-block:: python + + from fastNLP import Const + + dataset.rename_field('words', Const.INPUT) + dataset.rename_field('seq_len', Const.INPUT_LEN) + dataset.rename_field('target', Const.TARGET) + + print(Const.INPUT) + print(Const.INPUT_LEN) + print(Const.TARGET) + print(Const.OUTPUT) + + 输出结果为:: + + words + seq_len + target + pred + + 在给 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 改名后，我们还需要设置训练所需的输入和目标，这里使用的是 + :meth:`~fastNLP.DataSet.set_input` 和 :meth:`~fastNLP.DataSet.set_target` 两个函数。 + + .. code-block:: python + + #使用dataset的 set_input 和 set_target函数，告诉模型dataset中那些数据是输入，那些数据是标签（目标输出） + dataset.set_input(Const.INPUT, Const.INPUT_LEN) + dataset.set_target(Const.TARGET) + +数据集分割 + 除了修改 :mod:`~fastNLP.core.field` 之外，我们还可以对 :class:`~fastNLP.DataSet` 进行分割，以供训练、开发和测试使用。 + 下面这段代码展示了 :meth:`~fastNLP.DataSet.split` 的使用方法 + + .. code-block:: python + + train_dev_data, test_data = dataset.split(0.1) + train_data, dev_data = train_dev_data.split(0.1) + print(len(train_data), len(dev_data), len(test_data)) + + 输出结果为:: + + 9603 1067 1185 + +评价指标 + 训练模型需要提供一个评价指标。这里使用准确率做为评价指标。参数的 `命名规则` 跟上面类似。 + ``pred`` 参数对应的是模型的 forward 方法返回的 dict 中的一个 key 的名字。 + ``target`` 参数对应的是 :class:`~fastNLP.DataSet` 中作为标签的 :mod:`~fastNLP.core.field` 的名字。 + + .. code-block:: python + + from fastNLP import AccuracyMetric + + # metrics=AccuracyMetric() 在本例中与下面这行代码等价 + metrics=AccuracyMetric(pred=Const.OUTPUT, target=Const.TARGET) + +损失函数 + 训练模型需要提供一个损失函数 + ,fastNLP中提供了直接可以导入使用的四种loss，分别为： + * :class:`~fastNLP.CrossEntropyLoss`：包装了torch.nn.functional.cross_entropy()函数，返回交叉熵损失（可以运用于多分类场景） + * :class:`~fastNLP.BCELoss`：包装了torch.nn.functional.binary_cross_entropy()函数，返回二分类的交叉熵 + * :class:`~fastNLP.L1Loss`：包装了torch.nn.functional.l1_loss()函数，返回L1 损失 + * :class:`~fastNLP.NLLLoss`：包装了torch.nn.functional.nll_loss()函数，返回负对数似然损失 + + 下面提供了一个在分类问题中常用的交叉熵损失。注意它的 **初始化参数** 。 + ``pred`` 参数对应的是模型的 forward 方法返回的 dict 中的一个 key 的名字。 + ``target`` 参数对应的是 :class:`~fastNLP.DataSet` 中作为标签的 :mod:`~fastNLP.core.field` 的名字。 + 这里我们用 :class:`~fastNLP.Const` 来辅助命名，如果你自己编写模型中 forward 方法的返回值或 + 数据集中 :mod:`~fastNLP.core.field` 的名字与本例不同， 你可以把 ``pred`` 参数和 ``target`` 参数设定符合自己代码的值。 + + .. code-block:: python + + from fastNLP import CrossEntropyLoss + + # loss = CrossEntropyLoss() 在本例中与下面这行代码等价 + loss = CrossEntropyLoss(pred=Const.OUTPUT, target=Const.TARGET) + +优化器 + 定义模型运行的时候使用的优化器，可以使用fastNLP包装好的优化器： + + * :class:`~fastNLP.SGD` ：包装了torch.optim.SGD优化器 + * :class:`~fastNLP.Adam` ：包装了torch.optim.Adam优化器 + + 也可以直接使用torch.optim.Optimizer中的优化器，并在实例化 :class:`~fastNLP.Trainer` 类的时候传入优化器实参 + + .. code-block:: python + + import torch.optim as optim + from fastNLP import Adam + + #使用 torch.optim 定义优化器 + optimizer_1=optim.RMSprop(model_cnn.parameters(), lr=0.01, alpha=0.99, eps=1e-08, weight_decay=0, momentum=0, centered=False) + #使用fastNLP中包装的 Adam 定义优化器 + optimizer_2=Adam(lr=4e-3, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, model_params=model_cnn.parameters()) + +快速训练 + 现在我们可以导入 fastNLP 内置的文本分类模型 :class:`~fastNLP.models.CNNText` ，并使用 :class:`~fastNLP.Trainer` 进行训练， + 除了使用 :class:`~fastNLP.Trainer`进行训练，我们也可以通过使用 :class:`~fastNLP.DataSetIter` 来编写自己的训练过程，具体见 :doc:`/tutorials/tutorial_5_datasetiter` + + .. code-block:: python + + from fastNLP.models import CNNText + + #词嵌入的维度、训练的轮数和batch size + EMBED_DIM = 100 + N_EPOCHS = 10 + BATCH_SIZE = 16 + + #使用CNNText的时候第一个参数输入一个tuple,作为模型定义embedding的参数 + #还可以传入 kernel_nums, kernel_sizes, padding, dropout的自定义值 + model_cnn = CNNText((len(vocab),EMBED_DIM), num_classes=3, padding=2, dropout=0.1) + + #如果在定义trainer的时候没有传入optimizer参数，模型默认的优化器为torch.optim.Adam且learning rate为lr=4e-3 + #这里只使用了optimizer_1作为优化器输入，感兴趣可以尝试optimizer_2或者其他优化器作为输入 + #这里只使用了loss作为损失函数输入，感兴趣可以尝试其他损失函数输入 + trainer = Trainer(model=model_cnn, train_data=train_data, dev_data=dev_data, loss=loss, metrics=metrics, + optimizer=optimizer_1,n_epochs=N_EPOCHS, batch_size=BATCH_SIZE) + trainer.train() + + 训练过程的输出如下:: + + input fields after batch(if batch size is 2): + words: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2, 40]) + seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) + target fields after batch(if batch size is 2): + target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) + + training epochs started 2019-07-08-15-44-48 + Evaluation at Epoch 1/10. Step:601/6010. AccuracyMetric: acc=0.59044 + + Evaluation at Epoch 2/10. Step:1202/6010. AccuracyMetric: acc=0.599813 + + Evaluation at Epoch 3/10. Step:1803/6010. AccuracyMetric: acc=0.508903 + + Evaluation at Epoch 4/10. Step:2404/6010. AccuracyMetric: acc=0.596064 + + Evaluation at Epoch 5/10. Step:3005/6010. AccuracyMetric: acc=0.47985 + + Evaluation at Epoch 6/10. Step:3606/6010. AccuracyMetric: acc=0.589503 + + Evaluation at Epoch 7/10. Step:4207/6010. AccuracyMetric: acc=0.311153 + + Evaluation at Epoch 8/10. Step:4808/6010. AccuracyMetric: acc=0.549203 + + Evaluation at Epoch 9/10. Step:5409/6010. AccuracyMetric: acc=0.581068 + + Evaluation at Epoch 10/10. Step:6010/6010. AccuracyMetric: acc=0.523899 + + + In Epoch:2/Step:1202, got best dev performance:AccuracyMetric: acc=0.599813 + Reloaded the best model. + +快速测试 + 与 :class:`~fastNLP.Trainer` 对应，fastNLP 也提供了 :class:`~fastNLP.Tester` 用于快速测试，用法如下 + + .. code-block:: python + + from fastNLP import Tester + + tester = Tester(test_data, model_cnn, metrics=AccuracyMetric()) + tester.test() + + 训练过程输出如下:: + + [tester] + AccuracyMetric: acc=0.565401 diff --git a/docs/source/tutorials/tutorial_5_datasetiter.rst b/docs/source/tutorials/tutorial_5_datasetiter.rst new file mode 100644 index 00000000..23d26deb --- /dev/null +++ b/docs/source/tutorials/tutorial_5_datasetiter.rst @@ -0,0 +1,250 @@ +============================================================================== +动手实现一个文本分类器II-使用DataSetIter实现自定义训练过程 +============================================================================== + +我们使用和 :doc:`/user/quickstart` 中一样的任务来进行详细的介绍。给出一段评价性文字，预测其情感倾向是积极（label=1）、 +消极（label=0）还是中性（label=2），使用 :class:`~fastNLP.DataSetIter` 类来编写自己的训练过程。 +自己编写训练过程之前的内容与 :doc:`/tutorials/tutorial_4_loss_optimizer` 中的完全一样，如已经阅读过可以跳过。 + +-------------- +数据处理 +-------------- + +数据读入 + 我们可以使用 fastNLP :mod:`fastNLP.io` 模块中的 :class:`~fastNLP.io.SSTLoader` 类，轻松地读取SST数据集（数据来源：https://nlp.stanford.edu/sentiment/trainDevTestTrees_PTB.zip）。 + 这里的 dataset 是 fastNLP 中 :class:`~fastNLP.DataSet` 类的对象。 + + .. code-block:: python + + from fastNLP.io import SSTLoader + + loader = SSTLoader() + #这里的all.txt是下载好数据后train.txt、dev.txt、test.txt的组合 + dataset = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + print(dataset[0]) + + 输出数据如下:: + + {'words': ['It', "'s", 'a', 'lovely', 'film', 'with', 'lovely', 'performances', 'by', 'Buy', 'and', 'Accorsi', '.'] type=list, + 'target': positive type=str} + + 除了读取数据外，fastNLP 还提供了读取其它文件类型的 Loader 类、读取 Embedding的 Loader 等。详见 :doc:`/fastNLP.io` 。 + + +数据处理 + 我们使用 :class:`~fastNLP.DataSet` 类的 :meth:`~fastNLP.DataSet.apply` 方法将 ``target`` :mod:`~fastNLP.core.field` 转化为整数。 + + .. code-block:: python + + def label_to_int(x): + if x['target']=="positive": + return 1 + elif x['target']=="negative": + return 0 + else: + return 2 + + # 将label转为整数 + dataset.apply(lambda x: label_to_int(x), new_field_name='target') + + ``words`` 和 ``target`` 已经足够用于 :class:`~fastNLP.models.CNNText` 的训练了，但我们从其文档 + :class:`~fastNLP.models.CNNText` 中看到，在 :meth:`~fastNLP.models.CNNText.forward` 的时候，还可以传入可选参数 ``seq_len`` 。 + 所以，我们再使用 :meth:`~fastNLP.DataSet.apply_field` 方法增加一个名为 ``seq_len`` 的 :mod:`~fastNLP.core.field` 。 + + .. code-block:: python + + # 增加长度信息 + dataset.apply_field(lambda x: len(x), field_name='words', new_field_name='seq_len') + + 观察可知： :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 类似， + 但所传入的 `lambda` 函数是针对一个 :class:`~fastNLP.Instance` 中的一个 :mod:`~fastNLP.core.field` 的； + 而 :meth:`~fastNLP.DataSet.apply` 所传入的 `lambda` 函数是针对整个 :class:`~fastNLP.Instance` 的。 + + .. note:: + `lambda` 函数即匿名函数，是 Python 的重要特性。 ``lambda x: len(x)`` 和下面的这个函数的作用相同:: + + def func_lambda(x): + return len(x) + + 你也可以编写复杂的函数做为 :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 的参数 + +Vocabulary 的使用 + 我们再用 :class:`~fastNLP.Vocabulary` 类来统计数据中出现的单词，并使用 :meth:`~fastNLP.Vocabulary.index_dataset` + 将单词序列转化为训练可用的数字序列。 + + .. code-block:: python + + from fastNLP import Vocabulary + + # 使用Vocabulary类统计单词，并将单词序列转化为数字序列 + vocab = Vocabulary(min_freq=2).from_dataset(dataset, field_name='words') + vocab.index_dataset(dataset, field_name='words',new_field_name='words') + print(dataset[0]) + + 输出数据如下:: + + {'words': [27, 9, 6, 913, 16, 18, 913, 124, 31, 5715, 5, 1, 2] type=list, + 'target': 1 type=int, + 'seq_len': 13 type=int} + + +--------------------- +使用内置模型训练 +--------------------- + +内置模型的输入输出命名 + fastNLP内置了一些完整的神经网络模型，详见 :doc:`/fastNLP.models` , 我们使用其中的 :class:`~fastNLP.models.CNNText` 模型进行训练。 + 为了使用内置的 :class:`~fastNLP.models.CNNText`，我们必须修改 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 的名称。 + 在这个例子中模型输入 (forward方法的参数) 为 ``words`` 和 ``seq_len`` ; 预测输出为 ``pred`` ;标准答案为 ``target`` 。 + 具体的命名规范可以参考 :doc:`/fastNLP.core.const` 。 + + 如果不想查看文档，您也可以使用 :class:`~fastNLP.Const` 类进行命名。下面的代码展示了给 :class:`~fastNLP.DataSet` 中 + :mod:`~fastNLP.core.field` 改名的 :meth:`~fastNLP.DataSet.rename_field` 方法，以及 :class:`~fastNLP.Const` 类的使用方法。 + + .. code-block:: python + + from fastNLP import Const + + dataset.rename_field('words', Const.INPUT) + dataset.rename_field('seq_len', Const.INPUT_LEN) + dataset.rename_field('target', Const.TARGET) + + print(Const.INPUT) + print(Const.INPUT_LEN) + print(Const.TARGET) + print(Const.OUTPUT) + + 输出结果为:: + + words + seq_len + target + pred + + 在给 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 改名后，我们还需要设置训练所需的输入和目标，这里使用的是 + :meth:`~fastNLP.DataSet.set_input` 和 :meth:`~fastNLP.DataSet.set_target` 两个函数。 + + .. code-block:: python + + #使用dataset的 set_input 和 set_target函数，告诉模型dataset中那些数据是输入，那些数据是标签（目标输出） + dataset.set_input(Const.INPUT, Const.INPUT_LEN) + dataset.set_target(Const.TARGET) + +数据集分割 + 除了修改 :mod:`~fastNLP.core.field` 之外，我们还可以对 :class:`~fastNLP.DataSet` 进行分割，以供训练、开发和测试使用。 + 下面这段代码展示了 :meth:`~fastNLP.DataSet.split` 的使用方法 + + .. code-block:: python + + train_dev_data, test_data = dataset.split(0.1) + train_data, dev_data = train_dev_data.split(0.1) + print(len(train_data), len(dev_data), len(test_data)) + + 输出结果为:: + + 9603 1067 1185 + +评价指标 + 训练模型需要提供一个评价指标。这里使用准确率做为评价指标。参数的 `命名规则` 跟上面类似。 + ``pred`` 参数对应的是模型的 forward 方法返回的 dict 中的一个 key 的名字。 + ``target`` 参数对应的是 :class:`~fastNLP.DataSet` 中作为标签的 :mod:`~fastNLP.core.field` 的名字。 + + .. code-block:: python + + from fastNLP import AccuracyMetric + + # metrics=AccuracyMetric() 在本例中与下面这行代码等价 + metrics=AccuracyMetric(pred=Const.OUTPUT, target=Const.TARGET) + + +-------------------------- +自己编写训练过程 +-------------------------- + 如果你想用类似 PyTorch 的使用方法，自己编写训练过程，你可以参考下面这段代码。 + 其中使用了 fastNLP 提供的 :class:`~fastNLP.DataSetIter` 来获得小批量训练的小批量数据， + 使用 :class:`~fastNLP.BucketSampler` 做为 :class:`~fastNLP.DataSetIter` 的参数来选择采样的方式。 + +DataSetIter + fastNLP定义的 :class:`~fastNLP.DataSetIter` 类，用于定义一个batch，并实现batch的多种功能，在初始化时传入的参数有： + + * dataset: :class:`~fastNLP.DataSet` 对象, 数据集 + * batch_size: 取出的batch大小 + * sampler: 规定使用的 :class:`~fastNLP.Sampler` 若为 None, 使用 :class:`~fastNLP.RandomSampler` （Default: None） + * as_numpy: 若为 True, 输出batch为 `numpy.array`. 否则为 `torch.Tensor` （Default: False） + * prefetch: 若为 True使用多进程预先取出下一batch. （Default: False） + +sampler + fastNLP 实现的采样器有： + + * :class:`~fastNLP.BucketSampler` 可以随机地取出长度相似的元素 【初始化参数: num_buckets：bucket的数量； batch_size：batch大小； seq_len_field_name：dataset中对应序列长度的 :mod:`~fastNLP.core.field` 的名字】 + * SequentialSampler： 顺序取出元素的采样器【无初始化参数】 + * RandomSampler：随机化取元素的采样器【无初始化参数】 + + 以下代码使用BucketSampler作为 :class:`~fastNLP.DataSetIter` 初始化的输入，运用 :class:`~fastNLP.DataSetIter` 自己写训练程序 + + .. code-block:: python + + from fastNLP import BucketSampler + from fastNLP import DataSetIter + from fastNLP.models import CNNText + from fastNLP import Tester + import torch + import time + + embed_dim = 100 + model = CNNText((len(vocab),embed_dim), num_classes=3, padding=2, dropout=0.1) + + def train(epoch, data, devdata): + optimizer = torch.optim.Adam(model.parameters(), lr=0.001) + lossfunc = torch.nn.CrossEntropyLoss() + batch_size = 32 + + # 定义一个Batch，传入DataSet，规定batch_size和去batch的规则。 + # 顺序（Sequential），随机（Random），相似长度组成一个batch（Bucket） + train_sampler = BucketSampler(batch_size=batch_size, seq_len_field_name='seq_len') + train_batch = DataSetIter(batch_size=batch_size, dataset=data, sampler=train_sampler) + + start_time = time.time() + print("-"*5+"start training"+"-"*5) + for i in range(epoch): + loss_list = [] + for batch_x, batch_y in train_batch: + optimizer.zero_grad() + output = model(batch_x['words']) + loss = lossfunc(output['pred'], batch_y['target']) + loss.backward() + optimizer.step() + loss_list.append(loss.item()) + + #这里verbose如果为0，在调用Tester对象的test()函数时不输出任何信息，返回评估信息; 如果为1，打印出验证结果，返回评估信息 + #在调用过Tester对象的test()函数后，调用其_format_eval_results(res)函数，结构化输出验证结果 + tester_tmp = Tester(devdata, model, metrics=AccuracyMetric(), verbose=0) + res=tester_tmp.test() + + print('Epoch {:d} Avg Loss: {:.2f}'.format(i, sum(loss_list) / len(loss_list)),end=" ") + print(tester._format_eval_results(res),end=" ") + print('{:d}ms'.format(round((time.time()-start_time)*1000))) + loss_list.clear() + + train(10, train_data, dev_data) + #使用tester进行快速测试 + tester = Tester(test_data, model, metrics=AccuracyMetric()) + tester.test() + + 这段代码的输出如下:: + + -----start training----- + Epoch 0 Avg Loss: 1.09 AccuracyMetric: acc=0.480787 58989ms + Epoch 1 Avg Loss: 1.00 AccuracyMetric: acc=0.500469 118348ms + Epoch 2 Avg Loss: 0.93 AccuracyMetric: acc=0.536082 176220ms + Epoch 3 Avg Loss: 0.87 AccuracyMetric: acc=0.556701 236032ms + Epoch 4 Avg Loss: 0.78 AccuracyMetric: acc=0.562324 294351ms + Epoch 5 Avg Loss: 0.69 AccuracyMetric: acc=0.58388 353673ms + Epoch 6 Avg Loss: 0.60 AccuracyMetric: acc=0.574508 412106ms + Epoch 7 Avg Loss: 0.51 AccuracyMetric: acc=0.589503 471097ms + Epoch 8 Avg Loss: 0.44 AccuracyMetric: acc=0.581068 529174ms + Epoch 9 Avg Loss: 0.39 AccuracyMetric: acc=0.572634 586216ms + [tester] + AccuracyMetric: acc=0.527426 + + diff --git a/docs/source/tutorials/tutorial_6_seq_labeling.rst b/docs/source/tutorials/tutorial_6_seq_labeling.rst new file mode 100644 index 00000000..09a53cdc --- /dev/null +++ b/docs/source/tutorials/tutorial_6_seq_labeling.rst @@ -0,0 +1,114 @@ +===================== +快速实现序列标注模型 +===================== + +这一部分的内容主要展示如何使用fastNLP 实现序列标注任务。你可以使用fastNLP的各个组件快捷，方便地完成序列标注任务，达到出色的效果。 +在阅读这篇Tutorial前，希望你已经熟悉了fastNLP的基础使用，包括基本数据结构以及数据预处理，embedding的嵌入等，希望你对之前的教程有更进一步的掌握。 +我们将对CoNLL-03的英文数据集进行处理，展示如何完成命名实体标注任务整个训练的过程。 + +载入数据 +=================================== +fastNLP可以方便地载入各种类型的数据。同时，针对常见的数据集，我们已经预先实现了载入方法，其中包含CoNLL-03数据集。 +在设计dataloader时，以DataSetLoader为基类，可以改写并应用于其他数据集的载入。 + +.. code-block:: python + + class Conll2003DataLoader(DataSetLoader): + def __init__(self, task:str='ner', encoding_type:str='bioes'): + assert task in ('ner', 'pos', 'chunk') + index = {'ner':3, 'pos':1, 'chunk':2}[task] + #ConllLoader是fastNLP内置的类 + self._loader = ConllLoader(headers=['raw_words', 'target'], indexes=[0, index]) + self._tag_converters = None + if task in ('ner', 'chunk'): + #iob和iob2bioes会对tag进行统一，标准化 + self._tag_converters = [iob2] + if encoding_type == 'bioes': + self._tag_converters.append(iob2bioes) + + def load(self, path: str): + dataset = self._loader.load(path) + def convert_tag_schema(tags): + for converter in self._tag_converters: + tags = converter(tags) + return tags + if self._tag_converters: + #使用apply实现convert_tag_schema函数，实际上也支持匿名函数 + dataset.apply_field(convert_tag_schema, field_name=Const.TARGET, new_field_name=Const.TARGET) + return dataset + +输出数据格式如： + + {'raw_words': ['on', 'Friday', ':'] type=list, + 'target': ['O', 'O', 'O'] type=list}, + + +数据处理 +---------------------------- +我们进一步处理数据。将数据和词表封装在 :class:`~fastNLP.DataBundle` 类中。data是DataBundle的实例。 +我们输入模型的数据包括char embedding，以及word embedding。在数据处理部分，我们尝试完成词表的构建。 +使用fastNLP中的Vocabulary类来构建词表。 + +.. code-block:: python + + word_vocab = Vocabulary(min_freq=2) + word_vocab.from_dataset(data.datasets['train'], field_name=Const.INPUT) + word_vocab.index_dataset(*data.datasets.values(),field_name=Const.INPUT, new_field_name=Const.INPUT) + +处理后的data对象内部为： + + dataset + vocabs + dataset保存了train和test中的数据，并保存为dataset类型 + vocab保存了words，raw-words以及target的词表。 + +模型构建 +-------------------------------- +我们使用CNN-BILSTM-CRF模型完成这一任务。在网络构建方面，fastNLP的网络定义继承pytorch的 :class:`nn.Module` 类。 +自己可以按照pytorch的方式定义网络。需要注意的是命名。fastNLP的标准命名位于 :class:`~fastNLP.Const` 类。 + +模型的训练 +首先实例化模型，导入所需的char embedding以及word embedding。Embedding的载入可以参考教程。 +也可以查看 :mod:`~fastNLP.modules.encoder.embedding` 使用所需的embedding 载入方法。 +fastNLP将模型的训练过程封装在了 :class:`~fastnlp.trainer` 类中。 +根据不同的任务调整trainer中的参数即可。通常，一个trainer实例需要有：指定的训练数据集，模型，优化器，loss函数，评测指标，以及指定训练的epoch数，batch size等参数。 + +.. code-block:: python + + #实例化模型 + model = CNNBiLSTMCRF(word_embed, char_embed, hidden_size=200, num_layers=1, tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type) + #定义优化器 + optimizer = Adam(model.parameters(), lr=0.005) + #定义评估指标 + Metrics=SpanFPreRecMetric(tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type) + #实例化trainer + trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=optimizer, dev_data=data.datasets['test'], batch_size=10, metrics=Metrics,callbacks=callbacks, n_epochs=100) + #开始训练 + trainer.train() + +训练中会保存最优的参数配置。 +训练的结果如下： + +.. code-block:: python + + Evaluation on DataSet test: + SpanFPreRecMetric: f=0.727661, pre=0.732293, rec=0.723088 + Evaluation at Epoch 1/100. Step:1405/140500. SpanFPreRecMetric: f=0.727661, pre=0.732293, rec=0.723088 + + Evaluation on DataSet test: + SpanFPreRecMetric: f=0.784307, pre=0.779371, rec=0.789306 + Evaluation at Epoch 2/100. Step:2810/140500. SpanFPreRecMetric: f=0.784307, pre=0.779371, rec=0.789306 + + Evaluation on DataSet test: + SpanFPreRecMetric: f=0.810068, pre=0.811003, rec=0.809136 + Evaluation at Epoch 3/100. Step:4215/140500. SpanFPreRecMetric: f=0.810068, pre=0.811003, rec=0.809136 + + Evaluation on DataSet test: + SpanFPreRecMetric: f=0.829592, pre=0.84153, rec=0.817989 + Evaluation at Epoch 4/100. Step:5620/140500. SpanFPreRecMetric: f=0.829592, pre=0.84153, rec=0.817989 + + Evaluation on DataSet test: + SpanFPreRecMetric: f=0.828789, pre=0.837096, rec=0.820644 + Evaluation at Epoch 5/100. Step:7025/140500. SpanFPreRecMetric: f=0.828789, pre=0.837096, rec=0.820644 + + diff --git a/docs/source/tutorials/tutorial_7_modules_models.rst b/docs/source/tutorials/tutorial_7_modules_models.rst new file mode 100644 index 00000000..680d75fd --- /dev/null +++ b/docs/source/tutorials/tutorial_7_modules_models.rst @@ -0,0 +1,207 @@ +====================================== +使用Modules和Models快速搭建自定义模型 +====================================== + +:mod:`~fastNLP.modules` 和 :mod:`~fastNLP.models` 用于构建 fastNLP 所需的神经网络模型，它可以和 torch.nn 中的模型一起使用。 +下面我们会分三节介绍编写构建模型的具体方法。 + + +---------------------- +使用 models 中的模型 +---------------------- + +fastNLP 在 :mod:`~fastNLP.models` 模块中内置了如 :class:`~fastNLP.models.CNNText` 、 +:class:`~fastNLP.models.SeqLabeling` 等完整的模型，以供用户直接使用。 +以 :class:`~fastNLP.models.CNNText` 为例，我们看一个简单的文本分类的任务的实现过程。 + +首先是数据读入和处理部分，这里的代码和 :doc:`快速入门 ` 中一致。 + +.. code-block:: python + + from fastNLP.io import CSVLoader + from fastNLP import Vocabulary, CrossEntropyLoss, AccuracyMetric + + loader = CSVLoader(headers=('raw_sentence', 'label'), sep='\t') + dataset = loader.load("./sample_data/tutorial_sample_dataset.csv") + + dataset.apply(lambda x: x['raw_sentence'].lower(), new_field_name='sentence') + dataset.apply_field(lambda x: x.split(), field_name='sentence', new_field_name='words', is_input=True) + dataset.apply(lambda x: int(x['label']), new_field_name='target', is_target=True) + + train_dev_data, test_data = dataset.split(0.1) + train_data, dev_data = train_dev_data.split(0.1) + + vocab = Vocabulary(min_freq=2).from_dataset(train_data, field_name='words') + vocab.index_dataset(train_data, dev_data, test_data, field_name='words', new_field_name='words') + +然后我们从 :mod:`~fastNLP.models` 中导入 ``CNNText`` 模型，用它进行训练 + +.. code-block:: python + + from fastNLP.models import CNNText + from fastNLP import Trainer + + model_cnn = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) + + trainer = Trainer(model=model_cnn, train_data=train_data, dev_data=dev_data, + loss=CrossEntropyLoss(), metrics=AccuracyMetric()) + trainer.train() + +在 iPython 环境输入 `model_cnn` ，我们可以看到 ``model_cnn`` 的网络结构 + +.. parsed-literal:: + + CNNText( + (embed): Embedding( + 169, 50 + (dropout): Dropout(p=0.0) + ) + (conv_pool): ConvMaxpool( + (convs): ModuleList( + (0): Conv1d(50, 3, kernel_size=(3,), stride=(1,), padding=(2,)) + (1): Conv1d(50, 4, kernel_size=(4,), stride=(1,), padding=(2,)) + (2): Conv1d(50, 5, kernel_size=(5,), stride=(1,), padding=(2,)) + ) + ) + (dropout): Dropout(p=0.1) + (fc): Linear(in_features=12, out_features=5, bias=True) + ) + +FastNLP 中内置的 models 如下表所示，您可以点击具体的名称查看详细的 API： + +.. csv-table:: + :header: 名称, 介绍 + + :class:`~fastNLP.models.CNNText` , 使用 CNN 进行文本分类的模型 + :class:`~fastNLP.models.SeqLabeling` , 简单的序列标注模型 + :class:`~fastNLP.models.AdvSeqLabel` , 更大网络结构的序列标注模型 + :class:`~fastNLP.models.ESIM` , ESIM 模型的实现 + :class:`~fastNLP.models.StarTransEnc` , 带 word-embedding的Star-Transformer模 型 + :class:`~fastNLP.models.STSeqLabel` , 用于序列标注的 Star-Transformer 模型 + :class:`~fastNLP.models.STNLICls` ,用于自然语言推断 (NLI) 的 Star-Transformer 模型 + :class:`~fastNLP.models.STSeqCls` , 用于分类任务的 Star-Transformer 模型 + :class:`~fastNLP.models.BiaffineParser` , Biaffine 依存句法分析网络的实现 + +---------------------------- +使用 nn.torch 编写模型 +---------------------------- + +FastNLP 完全支持使用 pyTorch 编写的模型，但与 pyTorch 中编写模型的常见方法不同， +用于 fastNLP 的模型中 forward 函数需要返回一个字典，字典中至少需要包含 ``pred`` 这个字段。 + +下面是使用 pyTorch 中的 torch.nn 模块编写的文本分类，注意观察代码中标注的向量维度。 +由于 pyTorch 使用了约定俗成的维度设置，使得 forward 中需要多次处理维度顺序 + +.. code-block:: python + + import torch + import torch.nn as nn + + class LSTMText(nn.Module): + def __init__(self, vocab_size, embedding_dim, output_dim, hidden_dim=64, num_layers=2, dropout=0.5): + super().__init__() + + self.embedding = nn.Embedding(vocab_size, embedding_dim) + self.lstm = nn.LSTM(embedding_dim, hidden_dim, num_layers=num_layers, bidirectional=True, dropout=dropout) + self.fc = nn.Linear(hidden_dim * 2, output_dim) + self.dropout = nn.Dropout(dropout) + + def forward(self, words): + # (input) words : (batch_size, seq_len) + words = words.permute(1,0) + # words : (seq_len, batch_size) + + embedded = self.dropout(self.embedding(words)) + # embedded : (seq_len, batch_size, embedding_dim) + output, (hidden, cell) = self.lstm(embedded) + # output: (seq_len, batch_size, hidden_dim * 2) + # hidden: (num_layers * 2, batch_size, hidden_dim) + # cell: (num_layers * 2, batch_size, hidden_dim) + + hidden = torch.cat((hidden[-2, :, :], hidden[-1, :, :]), dim=1) + hidden = self.dropout(hidden) + # hidden: (batch_size, hidden_dim * 2) + + pred = self.fc(hidden.squeeze(0)) + # result: (batch_size, output_dim) + return {"pred":pred} + +我们同样可以在 iPython 环境中查看这个模型的网络结构 + +.. parsed-literal:: + + LSTMText( + (embedding): Embedding(169, 50) + (lstm): LSTM(50, 64, num_layers=2, dropout=0.5, bidirectional=True) + (fc): Linear(in_features=128, out_features=5, bias=True) + (dropout): Dropout(p=0.5) + ) + +---------------------------- +使用 modules 编写模型 +---------------------------- + +下面我们使用 :mod:`fastNLP.modules` 中的组件来构建同样的网络。由于 fastNLP 统一把 ``batch_size`` 放在第一维， +在编写代码的过程中会有一定的便利。 + +.. code-block:: python + + from fastNLP.modules import Embedding, LSTM, MLP + + class Model(nn.Module): + def __init__(self, vocab_size, embedding_dim, output_dim, hidden_dim=64, num_layers=2, dropout=0.5): + super().__init__() + + self.embedding = Embedding((vocab_size, embedding_dim)) + self.lstm = LSTM(embedding_dim, hidden_dim, num_layers=num_layers, bidirectional=True) + self.mlp = MLP([hidden_dim*2,output_dim], dropout=dropout) + + def forward(self, words): + embedded = self.embedding(words) + _,(hidden,_) = self.lstm(embedded) + pred = self.mlp(torch.cat((hidden[-1],hidden[-2]),dim=1)) + return {"pred":pred} + +我们自己编写模型的网络结构如下 + +.. parsed-literal:: + + Model( + (embedding): Embedding( + 169, 50 + (dropout): Dropout(p=0.0) + ) + (lstm): LSTM( + (lstm): LSTM(50, 64, num_layers=2, batch_first=True, bidirectional=True) + ) + (mlp): MLP( + (hiddens): ModuleList() + (output): Linear(in_features=128, out_features=5, bias=True) + (dropout): Dropout(p=0.5) + ) + ) + +FastNLP 中包含的各种模块如下表，您可以点击具体的名称查看详细的 API，也可以通过 :doc:`/fastNLP.modules` 进行了解。 + +.. csv-table:: + :header: 名称, 介绍 + + :class:`~fastNLP.modules.ConvolutionCharEncoder` , char级别的卷积 encoder + :class:`~fastNLP.modules.LSTMCharEncoder` , char级别基于LSTM的 encoder + :class:`~fastNLP.modules.ConvMaxpool` , 结合了Convolution和Max-Pooling于一体的模块 + :class:`~fastNLP.modules.LSTM` , LSTM模块, 轻量封装了PyTorch的LSTM + :class:`~fastNLP.modules.StarTransformer` , Star-Transformer 的encoder部分 + :class:`~fastNLP.modules.TransformerEncoder` , Transformer的encoder模块，不包含embedding层 + :class:`~fastNLP.modules.VarRNN` , Variational Dropout RNN 模块 + :class:`~fastNLP.modules.VarLSTM` , Variational Dropout LSTM 模块 + :class:`~fastNLP.modules.VarGRU` , Variational Dropout GRU 模块 + :class:`~fastNLP.modules.MaxPool` , Max-pooling模块 + :class:`~fastNLP.modules.MaxPoolWithMask` , 带mask矩阵的max pooling。在做 max-pooling的时候不会考虑mask值为0的位置。 + :class:`~fastNLP.modules.AvgPool` , Average-pooling模块 + :class:`~fastNLP.modules.AvgPoolWithMask` , 带mask矩阵的average pooling。在做 average-pooling的时候不会考虑mask值为0的位置。 + :class:`~fastNLP.modules.MultiHeadAttention` , MultiHead Attention 模块 + :class:`~fastNLP.modules.MLP` , 简单的多层感知器模块 + :class:`~fastNLP.modules.ConditionalRandomField` , 条件随机场模块 + :class:`~fastNLP.modules.viterbi_decode` , 给定一个特征矩阵以及转移分数矩阵，计算出最佳的路径以及对应的分数 （与 :class:`~fastNLP.modules.ConditionalRandomField` 配合使用） + :class:`~fastNLP.modules.allowed_transitions` , 给定一个id到label的映射表，返回所有可以跳转的列表（与 :class:`~fastNLP.modules.ConditionalRandomField` 配合使用） + :class:`~fastNLP.modules.TimestepDropout` , 简单包装过的Dropout 组件 diff --git a/docs/source/tutorials/tutorial_8_metrics.rst b/docs/source/tutorials/tutorial_8_metrics.rst new file mode 100644 index 00000000..0b4f86c8 --- /dev/null +++ b/docs/source/tutorials/tutorial_8_metrics.rst @@ -0,0 +1,121 @@ +=============================== +使用Metric快速评测你的模型 +=============================== + +在进行训练时，fastNLP提供了各种各样的 :mod:`~fastNLP.core.metrics` 。 +如 :doc:`/user/quickstart` 中所介绍的，:class:`~fastNLP.AccuracyMetric` 类的对象被直接传到 :class:`~fastNLP.Trainer` 中用于训练 + +.. code-block:: python + + from fastNLP import Trainer, CrossEntropyLoss, AccuracyMetric + + trainer = Trainer(model=model, train_data=train_data, dev_data=dev_data, + loss=CrossEntropyLoss(), metrics=AccuracyMetric()) + trainer.train() + +除了 :class:`~fastNLP.AccuracyMetric` 之外，:class:`~fastNLP.SpanFPreRecMetric` 也是一种非常见的评价指标， +例如在序列标注问题中，常以span的方式计算 F-measure, precision, recall。 + +另外，fastNLP 还实现了用于抽取式QA（如SQuAD）的metric :class:`~fastNLP.ExtractiveQAMetric`。 +用户可以参考下面这个表格，点击第一列查看各个 :mod:`~fastNLP.core.metrics` 的详细文档。 + +.. csv-table:: + :header: 名称, 介绍 + + :class:`~fastNLP.core.metrics.MetricBase` , 自定义metrics需继承的基类 + :class:`~fastNLP.core.metrics.AccuracyMetric` , 简单的正确率metric + :class:`~fastNLP.core.metrics.SpanFPreRecMetric` , "同时计算 F-measure, precision, recall 值的 metric" + :class:`~fastNLP.core.metrics.ExtractiveQAMetric` , 用于抽取式QA任务 的metric + +更多的 :mod:`~fastNLP.core.metrics` 正在被添加到 fastNLP 当中，敬请期待。 + +------------------------------ +定义自己的metrics +------------------------------ + +在定义自己的metrics类时需继承 fastNLP 的 :class:`~fastNLP.core.metrics.MetricBase`, +并覆盖写入 ``evaluate`` 和 ``get_metric`` 方法。 + + evaluate(xxx) 中传入一个批次的数据，将针对一个批次的预测结果做评价指标的累计 + + get_metric(xxx) 当所有数据处理完毕时调用该方法，它将根据 evaluate函数累计的评价指标统计量来计算最终的评价结果 + +以分类问题中，Accuracy计算为例，假设model的forward返回dict中包含 `pred` 这个key, 并且该key需要用于Accuracy:: + + class Model(nn.Module): + def __init__(xxx): + # do something + def forward(self, xxx): + # do something + return {'pred': pred, 'other_keys':xxx} # pred's shape: batch_size x num_classes + +假设dataset中 `label` 这个field是需要预测的值，并且该field被设置为了target +对应的AccMetric可以按如下的定义, version1, 只使用这一次:: + + class AccMetric(MetricBase): + def __init__(self): + super().__init__() + + # 根据你的情况自定义指标 + self.corr_num = 0 + self.total = 0 + + def evaluate(self, label, pred): # 这里的名称需要和dataset中target field与model返回的key是一样的，不然找不到对应的value + # dev或test时，每个batch结束会调用一次该方法，需要实现如何根据每个batch累加metric + self.total += label.size(0) + self.corr_num += label.eq(pred).sum().item() + + def get_metric(self, reset=True): # 在这里定义如何计算metric + acc = self.corr_num/self.total + if reset: # 是否清零以便重新计算 + self.corr_num = 0 + self.total = 0 + return {'acc': acc} # 需要返回一个dict，key为该metric的名称，该名称会显示到Trainer的progress bar中 + + +version2，如果需要复用Metric，比如下一次使用AccMetric时，dataset中目标field不叫label而叫y，或者model的输出不是pred:: + + class AccMetric(MetricBase): + def __init__(self, label=None, pred=None): + # 假设在另一场景使用时，目标field叫y，model给出的key为pred_y。则只需要在初始化AccMetric时， + # acc_metric = AccMetric(label='y', pred='pred_y')即可。 + # 当初始化为acc_metric = AccMetric()，即label=None, pred=None, fastNLP会直接使用'label', 'pred'作为key去索取对 + # 应的的值 + super().__init__() + self._init_param_map(label=label, pred=pred) # 该方法会注册label和pred. 仅需要注册evaluate()方法会用到的参数名即可 + # 如果没有注册该则效果与version1就是一样的 + + # 根据你的情况自定义指标 + self.corr_num = 0 + self.total = 0 + + def evaluate(self, label, pred): # 这里的参数名称需要和self._init_param_map()注册时一致。 + # dev或test时，每个batch结束会调用一次该方法，需要实现如何根据每个batch累加metric + self.total += label.size(0) + self.corr_num += label.eq(pred).sum().item() + + def get_metric(self, reset=True): # 在这里定义如何计算metric + acc = self.corr_num/self.total + if reset: # 是否清零以便重新计算 + self.corr_num = 0 + self.total = 0 + return {'acc': acc} # 需要返回一个dict，key为该metric的名称，该名称会显示到Trainer的progress bar中 + + +``MetricBase`` 将会在输入的字典 ``pred_dict`` 和 ``target_dict`` 中进行检查. +``pred_dict`` 是模型当中 ``forward()`` 函数或者 ``predict()`` 函数的返回值. +``target_dict`` 是DataSet当中的ground truth, 判定ground truth的条件是field的 ``is_target`` 被设置为True. + +``MetricBase`` 会进行以下的类型检测: + +1. self.evaluate当中是否有varargs, 这是不支持的. +2. self.evaluate当中所需要的参数是否既不在 ``pred_dict`` 也不在 ``target_dict`` . +3. self.evaluate当中所需要的参数是否既在 ``pred_dict`` 也在 ``target_dict`` . + +除此以外，在参数被传入self.evaluate以前，这个函数会检测 ``pred_dict`` 和 ``target_dict`` 当中没有被用到的参数 +如果kwargs是self.evaluate的参数，则不会检测 + + +self.evaluate将计算一个批次(batch)的评价指标，并累计。 没有返回值 +self.get_metric将统计当前的评价指标并返回评价结果, 返回值需要是一个dict, key是指标名称，value是指标的值 + diff --git a/docs/source/tutorials/tutorial_9_callback.rst b/docs/source/tutorials/tutorial_9_callback.rst new file mode 100644 index 00000000..8e2742bb --- /dev/null +++ b/docs/source/tutorials/tutorial_9_callback.rst @@ -0,0 +1,67 @@ +=================================================== +使用Callback自定义你的训练过程 +=================================================== + +在训练时，我们常常要使用trick来提高模型的性能（如调节学习率），或者要打印训练中的信息。 +这里我们提供Callback类，在Trainer中插入代码，完成一些自定义的操作。 + +我们使用和 :doc:`/user/quickstart` 中一样的任务来进行详细的介绍。 +给出一段评价性文字，预测其情感倾向是积极（label=1）、消极（label=0）还是中性（label=2），使用 :class:`~fastNLP.Trainer` 和 :class:`~fastNLP.Tester` 来进行快速训练和测试。 +关于数据处理，Loss和Optimizer的选择可以看其他教程，这里仅在训练时加入学习率衰减。 + +--------------------- +Callback的构建和使用 +--------------------- + +创建Callback + 我们可以继承fastNLP :class:`~fastNLP.Callback` 类来定义自己的Callback。 + 这里我们实现一个让学习率线性衰减的Callback。 + + .. code-block:: python + + import fastNLP + + class LRDecay(fastNLP.Callback): + def __init__(self): + super(MyCallback, self).__init__() + self.base_lrs = [] + self.delta = [] + + def on_train_begin(self): + # 初始化，仅训练开始时调用 + self.base_lrs = [pg['lr'] for pg in self.optimizer.param_groups] + self.delta = [float(lr) / self.n_epochs for lr in self.base_lrs] + + def on_epoch_end(self): + # 每个epoch结束时，更新学习率 + ep = self.epoch + lrs = [lr - d * ep for lr, d in zip(self.base_lrs, self.delta)] + self.change_lr(lrs) + + def change_lr(self, lrs): + for pg, lr in zip(self.optimizer.param_groups, lrs): + pg['lr'] = lr + + 这里，:class:`~fastNLP.Callback` 中所有以 ``on_`` 开头的类方法会在 :class:`~fastNLP.Trainer` 的训练中在特定时间调用。 + 如 on_train_begin() 会在训练开始时被调用，on_epoch_end() 会在每个 epoch 结束时调用。 + 具体有哪些类方法，参见文档 :class:`~fastNLP.Callback` 。 + + 另外，为了使用方便，可以在 :class:`~fastNLP.Callback` 内部访问 :class:`~fastNLP.Trainer` 中的属性，如 optimizer, epoch, step，分别对应训练时的优化器，当前epoch数，和当前的总step数。 + 具体可访问的属性，参见文档 :class:`~fastNLP.Callback` 。 + +使用Callback + 在定义好 :class:`~fastNLP.Callback` 之后，就能将它传入Trainer的 ``callbacks`` 参数，在实际训练时使用。 + + .. code-block:: python + + """ + 数据预处理，模型定义等等 + """ + + trainer = fastNLP.Trainer( + model=model, train_data=train_data, dev_data=dev_data, + optimizer=optimizer, metrics=metrics, + batch_size=10, n_epochs=100, + callbacks=[LRDecay()]) + + trainer.train() diff --git a/docs/source/user/docs_in_code.rst b/docs/source/user/docs_in_code.rst new file mode 100644 index 00000000..a0b9576f --- /dev/null +++ b/docs/source/user/docs_in_code.rst @@ -0,0 +1,3 @@ +=============== +在代码中写文档 +=============== \ No newline at end of file diff --git a/docs/source/user/example.rst b/docs/source/user/example.rst index 55588c79..70ebe628 100644 --- a/docs/source/user/example.rst +++ b/docs/source/user/example.rst @@ -20,7 +20,13 @@ 小标题4 ------------------- -参考 http://docutils.sourceforge.net/docs/user/rst/quickref.html +推荐使用大标题、小标题3和小标题4 + +官方文档 http://docutils.sourceforge.net/docs/user/rst/quickref.html + +`熟悉markdown的同学推荐参考这篇文章 `_ + +\<\>内表示的是链接地址，\<\>外的是显示到外面的文字 常见语法 ============ @@ -75,6 +81,7 @@ http://docutils.sf.net/ 孤立的网址会自动生成链接 不显示冒号的代码块 .. code-block:: python + :linenos: :emphasize-lines: 1,3 @@ -83,22 +90,67 @@ http://docutils.sf.net/ 孤立的网址会自动生成链接 print("有行号和高亮") 数学块 +========== .. math:: H_2O + Na = NaOH + H_2 \uparrow +复杂表格 +========== + ++------------------------+------------+----------+----------+ +| Header row, column 1 | Header 2 | Header 3 | Header 4 | +| (header rows optional) | | | | ++========================+============+==========+==========+ +| body row 1, column 1 | column 2 | column 3 | column 4 | ++------------------------+------------+----------+----------+ +| body row 2 | Cells may span columns. | ++------------------------+------------+---------------------+ +| body row 3 | Cells may | - Table cells | ++------------------------+ span rows. | - contain | +| body row 4 | | - body elements. | ++------------------------+------------+---------------------+ + +简易表格 +========== + +===== ===== ====== + Inputs Output +------------ ------ + A B A or B +===== ===== ====== +False False False +True True True +===== ===== ====== + +csv 表格 +============ + +.. csv-table:: + :header: sentence, target + + This is the first instance ., 0 + Second instance ., 1 + Third instance ., 1 + ..., ... + + + +[重要]各种链接 +=================== + +各种链接帮助我们连接到fastNLP文档的各个位置 -各种连接 -=========== +\<\>内表示的是链接地址，\<\>外的是显示到外面的文字 -:doc:`/user/with_fitlog` +:doc:`根据文件名链接 ` :mod:`~fastNLP.core.batch` :class:`~fastNLP.Batch` -~表示指显示最后一项 +~表示只显示最后一项 :meth:`fastNLP.DataSet.apply` diff --git a/docs/source/user/installation.rst b/docs/source/user/installation.rst index c218b3e1..42ea402c 100644 --- a/docs/source/user/installation.rst +++ b/docs/source/user/installation.rst @@ -7,10 +7,12 @@ fastNLP 依赖如下包:: - torch>=0.4.0 - numpy - tqdm - nltk + numpy>=1.14.2 + torch>=1.0.0 + tqdm>=4.28.1 + nltk>=3.4.1 + requests + spacy 其中torch的安装可能与操作系统及 CUDA 的版本相关，请参见 `PyTorch 官网 `_ 。 在依赖包安装完成的情况，您可以在命令行执行如下指令完成安装 @@ -18,3 +20,4 @@ fastNLP 依赖如下包:: .. code:: shell >>> pip install fastNLP + >>> python -m spacy download en diff --git a/docs/source/user/quickstart.rst b/docs/source/user/quickstart.rst index 12e541b7..b92645b0 100644 --- a/docs/source/user/quickstart.rst +++ b/docs/source/user/quickstart.rst @@ -121,4 +121,4 @@ In Epoch:6/Step:12, got best dev performance:AccuracyMetric: acc=0.8 Reloaded the best model. -这份教程只是简单地介绍了使用 fastNLP 工作的流程，具体的细节分析见 :doc:`/user/tutorial_one` +这份教程只是简单地介绍了使用 fastNLP 工作的流程，更多的教程分析见 :doc:`/user/tutorials` diff --git a/docs/source/user/tutorial_one.rst b/docs/source/user/tutorial_one.rst deleted file mode 100644 index 0c7be77d..00000000 --- a/docs/source/user/tutorial_one.rst +++ /dev/null @@ -1,371 +0,0 @@ -=============== -详细指南 -=============== - -我们使用和 :doc:`/user/quickstart` 中一样的任务来进行详细的介绍。给出一段文字，预测它的标签是0~4中的哪一个 -(数据来源 `kaggle `_ )。 - --------------- -数据处理 --------------- - -数据读入 - 我们可以使用 fastNLP :mod:`fastNLP.io` 模块中的 :class:`~fastNLP.io.CSVLoader` 类，轻松地从 csv 文件读取我们的数据。 - 这里的 dataset 是 fastNLP 中 :class:`~fastNLP.DataSet` 类的对象 - - .. code-block:: python - - from fastNLP.io import CSVLoader - - loader = CSVLoader(headers=('raw_sentence', 'label'), sep='\t') - dataset = loader.load("./sample_data/tutorial_sample_dataset.csv") - - 除了读取数据外，fastNLP 还提供了读取其它文件类型的 Loader 类、读取 Embedding的 Loader 等。详见 :doc:`/fastNLP.io` 。 - -Instance 和 DataSet - fastNLP 中的 :class:`~fastNLP.DataSet` 类对象类似于二维表格，它的每一列是一个 :mod:`~fastNLP.core.field` - 每一行是一个 :mod:`~fastNLP.core.instance` 。我们可以手动向数据集中添加 :class:`~fastNLP.Instance` 类的对象 - - .. code-block:: python - - from fastNLP import Instance - - dataset.append(Instance(raw_sentence='fake data', label='0')) - - 此时的 ``dataset[-1]`` 的值如下,可以看到，数据集中的每个数据包含 ``raw_sentence`` 和 ``label`` 两个 - :mod:`~fastNLP.core.field` ，他们的类型都是 ``str`` :: - - {'raw_sentence': fake data type=str, 'label': 0 type=str} - -field 的修改 - 我们使用 :class:`~fastNLP.DataSet` 类的 :meth:`~fastNLP.DataSet.apply` 方法将 ``raw_sentence`` 中字母变成小写，并将句子分词。 - 同时也将 ``label`` :mod:`~fastNLP.core.field` 转化为整数并改名为 ``target`` - - .. code-block:: python - - dataset.apply(lambda x: x['raw_sentence'].lower(), new_field_name='sentence') - dataset.apply_field(lambda x: x.split(), field_name='sentence', new_field_name='words') - dataset.apply(lambda x: int(x['label']), new_field_name='target') - - ``words`` 和 ``target`` 已经足够用于 :class:`~fastNLP.models.CNNText` 的训练了，但我们从其文档 - :class:`~fastNLP.models.CNNText` 中看到，在 :meth:`~fastNLP.models.CNNText.forward` 的时候，还可以传入可选参数 ``seq_len`` 。 - 所以，我们再使用 :meth:`~fastNLP.DataSet.apply_field` 方法增加一个名为 ``seq_len`` 的 :mod:`~fastNLP.core.field` 。 - - .. code-block:: python - - dataset.apply_field(lambda x: len(x), field_name='words', new_field_name='seq_len') - - 观察可知： :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 类似， - 但所传入的 `lambda` 函数是针对一个 :class:`~fastNLP.Instance` 中的一个 :mod:`~fastNLP.core.field` 的； - 而 :meth:`~fastNLP.DataSet.apply` 所传入的 `lambda` 函数是针对整个 :class:`~fastNLP.Instance` 的。 - - .. note:: - `lambda` 函数即匿名函数，是 Python 的重要特性。 ``lambda x: len(x)`` 和下面的这个函数的作用相同:: - - def func_lambda(x): - return len(x) - - 你也可以编写复杂的函数做为 :meth:`~fastNLP.DataSet.apply_field` 与 :meth:`~fastNLP.DataSet.apply` 的参数 - -Vocabulary 的使用 - 我们再用 :class:`~fastNLP.Vocabulary` 类来统计数据中出现的单词，并使用 :meth:`~fastNLP.Vocabularyindex_dataset` - 将单词序列转化为训练可用的数字序列。 - - .. code-block:: python - - from fastNLP import Vocabulary - - vocab = Vocabulary(min_freq=2).from_dataset(dataset, field_name='words') - vocab.index_dataset(dataset, field_name='words',new_field_name='words') - -数据集分割 - 除了修改 :mod:`~fastNLP.core.field` 之外，我们还可以对 :class:`~fastNLP.DataSet` 进行分割，以供训练、开发和测试使用。 - 下面这段代码展示了 :meth:`~fastNLP.DataSet.split` 的使用方法（但实际应该放在后面两段改名和设置输入的代码之后） - - .. code-block:: python - - train_dev_data, test_data = dataset.split(0.1) - train_data, dev_data = train_dev_data.split(0.1) - len(train_data), len(dev_data), len(test_data) - ---------------------- -使用内置模型训练 ---------------------- - -内置模型的输入输出命名 - fastNLP内置了一些完整的神经网络模型，详见 :doc:`/fastNLP.models` , 我们使用其中的 :class:`~fastNLP.models.CNNText` 模型进行训练。 - 为了使用内置的 :class:`~fastNLP.models.CNNText`，我们必须修改 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 的名称。 - 在这个例子中模型输入 (forward方法的参数) 为 ``words`` 和 ``seq_len`` ; 预测输出为 ``pred`` ;标准答案为 ``target`` 。 - 具体的命名规范可以参考 :doc:`/fastNLP.core.const` 。 - - 如果不想查看文档，您也可以使用 :class:`~fastNLP.Const` 类进行命名。下面的代码展示了给 :class:`~fastNLP.DataSet` 中 - :mod:`~fastNLP.core.field` 改名的 :meth:`~fastNLP.DataSet.rename_field` 方法，以及 :class:`~fastNLP.Const` 类的使用方法。 - - .. code-block:: python - - from fastNLP import Const - - dataset.rename_field('words', Const.INPUT) - dataset.rename_field('seq_len', Const.INPUT_LEN) - dataset.rename_field('target', Const.TARGET) - - 在给 :class:`~fastNLP.DataSet` 中 :mod:`~fastNLP.core.field` 改名后，我们还需要设置训练所需的输入和目标，这里使用的是 - :meth:`~fastNLP.DataSet.set_input` 和 :meth:`~fastNLP.DataSet.set_target` 两个函数。 - - .. code-block:: python - - dataset.set_input(Const.INPUT, Const.INPUT_LEN) - dataset.set_target(Const.TARGET) - -快速训练 - 现在我们可以导入 fastNLP 内置的文本分类模型 :class:`~fastNLP.models.CNNText` ，并使用 :class:`~fastNLP.Trainer` 进行训练了 - （其中 ``loss`` 和 ``metrics`` 的定义，我们将在后续两段代码中给出）。 - - .. code-block:: python - - from fastNLP.models import CNNText - from fastNLP import Trainer - - model = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) - - trainer = Trainer(model=model_cnn, train_data=train_data, dev_data=dev_data, - loss=loss, metrics=metrics) - trainer.train() - - 训练过程的输出如下:: - - input fields after batch(if batch size is 2): - words: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2, 26]) - target fields after batch(if batch size is 2): - target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) - - training epochs started 2019-05-09-10-59-39 - Evaluation at Epoch 1/10. Step:2/20. AccuracyMetric: acc=0.333333 - - Evaluation at Epoch 2/10. Step:4/20. AccuracyMetric: acc=0.533333 - - Evaluation at Epoch 3/10. Step:6/20. AccuracyMetric: acc=0.533333 - - Evaluation at Epoch 4/10. Step:8/20. AccuracyMetric: acc=0.533333 - - Evaluation at Epoch 5/10. Step:10/20. AccuracyMetric: acc=0.6 - - Evaluation at Epoch 6/10. Step:12/20. AccuracyMetric: acc=0.8 - - Evaluation at Epoch 7/10. Step:14/20. AccuracyMetric: acc=0.8 - - Evaluation at Epoch 8/10. Step:16/20. AccuracyMetric: acc=0.733333 - - Evaluation at Epoch 9/10. Step:18/20. AccuracyMetric: acc=0.733333 - - Evaluation at Epoch 10/10. Step:20/20. AccuracyMetric: acc=0.733333 - - - In Epoch:6/Step:12, got best dev performance:AccuracyMetric: acc=0.8 - Reloaded the best model. - -损失函数 - 训练模型需要提供一个损失函数, 下面提供了一个在分类问题中常用的交叉熵损失。注意它的 **初始化参数** 。 - ``pred`` 参数对应的是模型的 forward 方法返回的 dict 中的一个 key 的名字。 - ``target`` 参数对应的是 :class:`~fastNLP.DataSet` 中作为标签的 :mod:`~fastNLP.core.field` 的名字。 - 这里我们用 :class:`~fastNLP.Const` 来辅助命名，如果你自己编写模型中 forward 方法的返回值或 - 数据集中 :mod:`~fastNLP.core.field` 的名字与本例不同， 你可以把 ``pred`` 参数和 ``target`` 参数设定符合自己代码的值。 - - .. code-block:: python - - from fastNLP import CrossEntropyLoss - - # loss = CrossEntropyLoss() 在本例中与下面这行代码等价 - loss = CrossEntropyLoss(pred=Const.OUTPUT, target=Const.TARGET) - -评价指标 - 训练模型需要提供一个评价指标。这里使用准确率做为评价指标。参数的 `命名规则` 跟上面类似。 - ``pred`` 参数对应的是模型的 forward 方法返回的 dict 中的一个 key 的名字。 - ``target`` 参数对应的是 :class:`~fastNLP.DataSet` 中作为标签的 :mod:`~fastNLP.core.field` 的名字。 - - .. code-block:: python - - from fastNLP import AccuracyMetric - - # metrics=AccuracyMetric() 在本例中与下面这行代码等价 - metrics=AccuracyMetric(pred=Const.OUTPUT, target=Const.TARGET) - -快速测试 - 与 :class:`~fastNLP.Trainer` 对应，fastNLP 也提供了 :class:`~fastNLP.Tester` 用于快速测试，用法如下 - - .. code-block:: python - - from fastNLP import Tester - - tester = Tester(test_data, model_cnn, metrics=AccuracyMetric()) - tester.test() - ---------------------- -编写自己的模型 ---------------------- - -因为 fastNLP 是基于 `PyTorch `_ 开发的框架，所以我们可以基于 PyTorch 模型编写自己的神经网络模型。 -与标准的 PyTorch 模型不同，fastNLP 模型中 forward 方法返回的是一个字典，字典中至少需要包含 "pred" 这个字段。 -而 forward 方法的参数名称必须与 :class:`~fastNLP.DataSet` 中用 :meth:`~fastNLP.DataSet.set_input` 设定的名称一致。 -模型定义的代码如下: - -.. code-block:: python - - import torch - import torch.nn as nn - - class LSTMText(nn.Module): - def __init__(self, vocab_size, embedding_dim, output_dim, hidden_dim=64, num_layers=2, dropout=0.5): - super().__init__() - - self.embedding = nn.Embedding(vocab_size, embedding_dim) - self.lstm = nn.LSTM(embedding_dim, hidden_dim, num_layers=num_layers, bidirectional=True, dropout=dropout) - self.fc = nn.Linear(hidden_dim * 2, output_dim) - self.dropout = nn.Dropout(dropout) - - def forward(self, words): - # (input) words : (batch_size, seq_len) - words = words.permute(1,0) - # words : (seq_len, batch_size) - - embedded = self.dropout(self.embedding(words)) - # embedded : (seq_len, batch_size, embedding_dim) - output, (hidden, cell) = self.lstm(embedded) - # output: (seq_len, batch_size, hidden_dim * 2) - # hidden: (num_layers * 2, batch_size, hidden_dim) - # cell: (num_layers * 2, batch_size, hidden_dim) - - hidden = torch.cat((hidden[-2, :, :], hidden[-1, :, :]), dim=1) - hidden = self.dropout(hidden) - # hidden: (batch_size, hidden_dim * 2) - - pred = self.fc(hidden.squeeze(0)) - # result: (batch_size, output_dim) - return {"pred":pred} - -模型的使用方法与内置模型 :class:`~fastNLP.models.CNNText` 一致 - -.. code-block:: python - - model_lstm = LSTMText(len(vocab),50,5) - - trainer = Trainer(model=model_lstm, train_data=train_data, dev_data=dev_data, - loss=loss, metrics=metrics) - trainer.train() - - tester = Tester(test_data, model_lstm, metrics=AccuracyMetric()) - tester.test() - -.. todo:: - 使用 :doc:`/fastNLP.modules` 编写模型 - --------------------------- -自己编写训练过程 --------------------------- - -如果你想用类似 PyTorch 的使用方法，自己编写训练过程，你可以参考下面这段代码。其中使用了 fastNLP 提供的 :class:`~fastNLP.Batch` -来获得小批量训练的小批量数据，使用 :class:`~fastNLP.BucketSampler` 做为 :class:`~fastNLP.Batch` 的参数来选择采样的方式。 -这段代码中使用了 PyTorch 的 `torch.optim.Adam` 优化器 和 `torch.nn.CrossEntropyLoss` 损失函数，并自己计算了正确率 - -.. code-block:: python - - from fastNLP import BucketSampler - from fastNLP import Batch - import torch - import time - - model = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) - - def train(epoch, data): - optim = torch.optim.Adam(model.parameters(), lr=0.001) - lossfunc = torch.nn.CrossEntropyLoss() - batch_size = 32 - - train_sampler = BucketSampler(batch_size=batch_size, seq_len_field_name='seq_len') - train_batch = Batch(batch_size=batch_size, dataset=data, sampler=train_sampler) - - start_time = time.time() - for i in range(epoch): - loss_list = [] - for batch_x, batch_y in train_batch: - optim.zero_grad() - output = model(batch_x['words']) - loss = lossfunc(output['pred'], batch_y['target']) - loss.backward() - optim.step() - loss_list.append(loss.item()) - print('Epoch {:d} Avg Loss: {:.2f}'.format(i, sum(loss_list) / len(loss_list)),end=" ") - print('{:d}ms'.format(round((time.time()-start_time)*1000))) - loss_list.clear() - - train(10, train_data) - - tester = Tester(test_data, model, metrics=AccuracyMetric()) - tester.test() - -这段代码的输出如下:: - - Epoch 0 Avg Loss: 2.76 17ms - Epoch 1 Avg Loss: 2.55 29ms - Epoch 2 Avg Loss: 2.37 41ms - Epoch 3 Avg Loss: 2.30 53ms - Epoch 4 Avg Loss: 2.12 65ms - Epoch 5 Avg Loss: 2.16 76ms - Epoch 6 Avg Loss: 1.88 88ms - Epoch 7 Avg Loss: 1.84 99ms - Epoch 8 Avg Loss: 1.71 111ms - Epoch 9 Avg Loss: 1.62 122ms - [tester] - AccuracyMetric: acc=0.142857 - ----------------------------------- -使用 Callback 增强 Trainer ----------------------------------- - -如果你不想自己实现繁琐的训练过程，只希望在训练过程中实现一些自己的功能（比如：输出从训练开始到当前 batch 结束的总时间）， -你可以使用 fastNLP 提供的 :class:`~fastNLP.Callback` 类。下面的例子中，我们继承 :class:`~fastNLP.Callback` 类实现了这个功能。 - -.. code-block:: python - - from fastNLP import Callback - - start_time = time.time() - - class MyCallback(Callback): - def on_epoch_end(self): - print('Sum Time: {:d}ms\n\n'.format(round((time.time()-start_time)*1000))) - - - model = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) - trainer = Trainer(model=model, train_data=train_data, dev_data=dev_data, - loss=CrossEntropyLoss(), metrics=AccuracyMetric(), callbacks=[MyCallback()]) - trainer.train() - -训练输出如下:: - - input fields after batch(if batch size is 2): - words: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2, 16]) - seq_len: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) - target fields after batch(if batch size is 2): - target: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) - - training epochs started 2019-05-12-21-38-40 - Evaluation at Epoch 1/10. Step:2/20. AccuracyMetric: acc=0.285714 - - Sum Time: 51ms - - - ………………………… - - - Evaluation at Epoch 10/10. Step:20/20. AccuracyMetric: acc=0.857143 - - Sum Time: 212ms - - - - In Epoch:10/Step:20, got best dev performance:AccuracyMetric: acc=0.857143 - Reloaded the best model. - -这个例子只是介绍了 :class:`~fastNLP.Callback` 类的使用方法。实际应用（比如：负采样、Learning Rate Decay、Early Stop 等）中 -很多功能已经被 fastNLP 实现了。你可以直接 import 它们使用，详细请查看文档 :doc:`/fastNLP.core.callback` 。 \ No newline at end of file diff --git a/docs/source/user/tutorials.rst b/docs/source/user/tutorials.rst new file mode 100644 index 00000000..196f9c29 --- /dev/null +++ b/docs/source/user/tutorials.rst @@ -0,0 +1,20 @@ +======================== +fastNLP 详细使用教程 +======================== + +这里是更详细的使用教程。对于大部分的用户，我们建议你从第一篇开始顺序阅读；如果你只想了解其中的一部分，也可以进行选读。 + +.. toctree:: + :maxdepth: 1 + + 使用DataSet预处理文本 + 使用DataSetLoader加载数据集 + 使用Embedding模块将文本转成向量 + 动手实现一个文本分类器I-使用Trainer和Tester快速训练和测试 + 动手实现一个文本分类器II-使用DataSetIter实现自定义训练过程 + 快速实现序列标注模型 + 使用Modules和Models快速搭建自定义模型 + 使用Metric快速评测你的模型 + 使用Callback自定义你的训练过程 + 使用fitlog 辅助 fastNLP 进行科研 + diff --git a/fastNLP/__init__.py b/fastNLP/__init__.py index 12d421a2..ec192568 100644 --- a/fastNLP/__init__.py +++ b/fastNLP/__init__.py @@ -1,11 +1,12 @@ """ -fastNLP 由 :mod:`~fastNLP.core` 、 :mod:`~fastNLP.io` 、:mod:`~fastNLP.modules`、:mod:`~fastNLP.models` -等子模块组成，你可以点进去查看每个模块的文档。 +fastNLP 由 :mod:`~fastNLP.core` 、 :mod:`~fastNLP.io` 、:mod:`~fastNLP.embeddings` 、 :mod:`~fastNLP.modules`、 +:mod:`~fastNLP.models` 等子模块组成，你可以查看每个模块的文档。 - :mod:`~fastNLP.core` 是fastNLP 的核心模块，包括 DataSet、 Trainer、 Tester 等组件。详见文档 :doc:`/fastNLP.core` - :mod:`~fastNLP.io` 是实现输入输出的模块，包括了数据集的读取，模型的存取等功能。详见文档 :doc:`/fastNLP.io` +- :mod:`~fastNLP.embeddings` 提供用于构建复杂网络模型所需的各种embedding。详见文档 :doc:`/fastNLP.embeddings` - :mod:`~fastNLP.modules` 包含了用于搭建神经网络模型的诸多组件，可以帮助用户快速搭建自己所需的网络。详见文档 :doc:`/fastNLP.modules` -- :mod:`~fastNLP.models` 包含了一些使用 fastNLP 实现的完整网络模型，包括CNNText、SeqLabeling等常见模型。详见文档 :doc:`/fastNLP.models` +- :mod:`~fastNLP.models` 包含了一些使用 fastNLP 实现的完整网络模型，包括 :class:`~fastNLP.models.CNNText` 、 :class:`~fastNLP.models.SeqLabeling` 等常见模型。详见文档 :doc:`fastNLP.models` fastNLP 中最常用的组件可以直接从 fastNLP 包中 import ，他们的文档如下： """ @@ -56,9 +57,10 @@ __all__ = [ "cache_results" ] -__version__ = '0.4.0' +__version__ = '0.4.5' from .core import * from . import models from . import modules +from . import embeddings from .io import data_loader diff --git a/fastNLP/core/__init__.py b/fastNLP/core/__init__.py index efc83017..b246c6a0 100644 --- a/fastNLP/core/__init__.py +++ b/fastNLP/core/__init__.py @@ -1,17 +1,15 @@ """ core 模块里实现了 fastNLP 的核心框架，常用的功能都可以从 fastNLP 包中直接 import。当然你也同样可以从 core 模块的子模块中 import， -例如 Batch 组件有两种 import 的方式:: +例如 :class:`~fastNLP.DataSetIter` 组件有两种 import 的方式:: # 直接从 fastNLP 中 import - from fastNLP import Batch + from fastNLP import DataSetIter - # 从 core 模块的子模块 batch 中 import - from fastNLP.core.batch import Batch + # 从 core 模块的子模块 batch 中 import DataSetIter + from fastNLP.core.batch import DataSetIter 对于常用的功能，你只需要在 :doc:`fastNLP` 中查看即可。如果想了解各个子模块的具体作用，您可以在下面找到每个子模块的具体文档。 -.. todo:: - 介绍core 的子模块的分工，好像必要性不大 """ from .batch import DataSetIter, BatchIter, TorchLoaderIter diff --git a/fastNLP/core/_parallel_utils.py b/fastNLP/core/_parallel_utils.py index 4a7757d3..6b24d9f9 100644 --- a/fastNLP/core/_parallel_utils.py +++ b/fastNLP/core/_parallel_utils.py @@ -1,6 +1,7 @@ import threading import torch +from torch import nn from torch.nn.parallel.parallel_apply import get_a_var from torch.nn.parallel.scatter_gather import scatter_kwargs, gather @@ -86,3 +87,16 @@ def _data_parallel_wrapper(func_name, device_ids, output_device): outputs = parallel_apply(replicas, func_name, inputs, kwargs, device_ids[:len(replicas)]) return gather(outputs, output_device) return wrapper + + +def _model_contains_inner_module(model): + """ + + :param nn.Module model: 模型文件，判断是否内部包含model.module, 多用于check模型是否是nn.DataParallel, + nn.parallel.DistributedDataParallel。主要是在做形参匹配的时候需要使用最内部的model的function。 + :return: bool + """ + if isinstance(model, nn.Module): + if isinstance(model, (nn.DataParallel, nn.parallel.DistributedDataParallel)): + return True + return False \ No newline at end of file diff --git a/fastNLP/core/batch.py b/fastNLP/core/batch.py index 2d8c1a80..538f583a 100644 --- a/fastNLP/core/batch.py +++ b/fastNLP/core/batch.py @@ -1,18 +1,17 @@ """ -batch 模块实现了 fastNLP 所需的 Batch 类。 +batch 模块实现了 fastNLP 所需的 :class:`~fastNLP.core.batch.DataSetIter` 类。 """ __all__ = [ + "BatchIter", "DataSetIter", "TorchLoaderIter", ] import atexit -from queue import Empty, Full import numpy as np import torch -import torch.multiprocessing as mp import torch.utils.data from numbers import Number @@ -94,9 +93,13 @@ class DataSetGetter: class SamplerAdapter(torch.utils.data.Sampler): def __init__(self, sampler, dataset): + super().__init__(dataset) self.sampler = sampler self.dataset = dataset + def __len__(self): + return len(self.dataset) + def __iter__(self): return iter(self.sampler(self.dataset)) @@ -166,15 +169,19 @@ class DataSetIter(BatchIter): timeout=0, worker_init_fn=None): super().__init__() assert isinstance(dataset, DataSet) - sampler = SamplerAdapter(sampler=sampler or SequentialSampler(), dataset=dataset) + if not isinstance(sampler, torch.utils.data.Sampler): + self.sampler = SamplerAdapter(sampler=sampler or SequentialSampler(), dataset=dataset) + else: + self.sampler = sampler dataset = DataSetGetter(dataset, as_numpy) collate_fn = dataset.collate_fn if hasattr(dataset, 'collate_fn') else None self.dataiter = torch.utils.data.DataLoader( - dataset=dataset, batch_size=batch_size, sampler=sampler, + dataset=dataset, batch_size=batch_size, sampler=self.sampler, collate_fn=collate_fn, num_workers=num_workers, pin_memory=pin_memory, drop_last=drop_last, timeout=timeout, worker_init_fn=worker_init_fn) - self.num_batches = self.get_num_batches(len(dataset), batch_size, drop_last) + # 以sampler的数量为准，因为DistributedSampler的时候每个进程上并不是所有的数据都用上了 + self.num_batches = self.get_num_batches(len(self.dataiter.sampler), batch_size, drop_last) self.batch_size = batch_size @@ -183,7 +190,7 @@ class TorchLoaderIter(BatchIter): super().__init__() assert isinstance(dataset, torch.utils.data.DataLoader) self.dataiter = dataset - self.num_batches = self.get_num_batches(len(dataset), dataset.batch_size, dataset.drop_last) + self.num_batches = self.get_num_batches(len(dataset.sampler), dataset.batch_size, dataset.drop_last) self.batch_size = dataset.batch_size diff --git a/fastNLP/core/callback.py b/fastNLP/core/callback.py index bbe2f325..1cc5d53b 100644 --- a/fastNLP/core/callback.py +++ b/fastNLP/core/callback.py @@ -2,11 +2,11 @@ r""" callback模块实现了 fastNLP 中的许多 callback 类，用于增强 :class:`~fastNLP.Trainer` 类。 虽然Trainer本身已经集成了一些功能，但仍然不足以囊括训练过程中可能需要到的功能， -比如负采样，learning rate decay, Early Stop等。 -为了解决这个问题fastNLP引入了callback的机制，Callback 是一种在Trainer训练过程中特定阶段会运行的函数集合。 -关于Trainer的详细文档，请参见 :doc:`trainer 模块` +比如负采样，learning rate decay 和 early stop等。 +为了解决这个问题，fastNLP引入了callback的机制，:class:`~fastNLP.Callback` 是一种在Trainer训练过程中特定阶段会运行的函数集合。 +关于 :class:`~fastNLP.Trainer` 的详细文档，请参见 :doc:`trainer 模块` -我们将 :meth:`~fastNLP.Train.train` 这个函数内部分为以下的阶段，在对应阶段会触发相应的调用:: +我们将 :meth:`~fastNLP.Trainer.train` 这个函数内部分为以下的阶段，在对应阶段会触发相应的调用:: callback.on_train_begin() # 开始进行训练 for i in range(1, n_epochs+1): @@ -31,8 +31,8 @@ callback模块实现了 fastNLP 中的许多 callback 类，用于增强 :class: callback.on_train_end() # 训练结束 callback.on_exception() # 这是一个特殊的步骤，在训练过程中遭遇exception会跳转到这里。 -如下面的例子所示，我们可以使用内置的 callback 类，或者继承 :class:`~fastNLP.core.callback.Callback` -定义自己的 callback 类:: +如下面的例子所示，我们可以使用内置的 callback 组件，或者继承 :class:`~fastNLP.core.callback.Callback` +定义自己的 callback 组件:: from fastNLP import Callback, EarlyStopCallback, Trainer, CrossEntropyLoss, AccuracyMetric from fastNLP.models import CNNText @@ -79,6 +79,7 @@ except: from ..io.model_io import ModelSaver, ModelLoader from .dataset import DataSet from .tester import Tester +import logging try: import fitlog @@ -100,7 +101,8 @@ class Callback(object): def __init__(self): super(Callback, self).__init__() self._trainer = None # 在Trainer内部被重新赋值 - + self._disabled = False + @property def trainer(self): """ @@ -158,7 +160,19 @@ class Callback(object): def batch_per_epoch(self): """每个epoch一共有多少个batch，只有在on_epoch_begin之后才能调用该属性。""" return self._trainer.batch_per_epoch - + + @property + def is_master(self): + return self._trainer.is_master() + + @property + def disabled(self): + return self._disabled + + @property + def logger(self): + return getattr(self._trainer, 'logger', logging) + def on_train_begin(self): """ 在Train过程开始之前调用。 @@ -250,6 +264,14 @@ class Callback(object): :return: """ pass + + def on_validation(self): + """ + 如果Trainer中设置了验证，则会在每次需要验证时调用该函数 + + :return: + """ + pass def on_epoch_end(self): """ @@ -281,6 +303,8 @@ def _transfer(func): def wrapper(manager, *arg): returns = [] for callback in manager.callbacks: + if callback.disabled: + continue returns.append(getattr(callback, func.__name__)(*arg)) return returns @@ -297,22 +321,28 @@ class CallbackManager(Callback): """ super(CallbackManager, self).__init__() # set attribute of trainer environment - + self._env = env self.callbacks = [] - if callbacks is not None: - if isinstance(callbacks, list): - if all([isinstance(cb, Callback) for cb in callbacks]) is True: - self.callbacks.extend(callbacks) - else: - obj = [not isinstance(cb, Callback) for cb in callbacks][0] - raise TypeError(f"Expect sub-classes of Callback. Got {type(obj)}") + if callbacks: + self.callbacks = self.prepare_callbacks(callbacks) + + def prepare_callbacks(self, callbacks): + if not callbacks: + return [] + if isinstance(callbacks, list): + if all([isinstance(cb, Callback) for cb in callbacks]) is True: + pass else: - raise TypeError(f"Expect callbacks in CallbackManager(callbacks) to be list. Got {type(callbacks)}.") - - for env_name, env_val in env.items(): - for callback in self.callbacks: + obj = [not isinstance(cb, Callback) for cb in callbacks][0] + raise TypeError(f"Expect sub-classes of Callback. Got {type(obj)}") + else: + raise TypeError(f"Expect callbacks in CallbackManager(callbacks) to be list. Got {type(callbacks)}.") + + for env_name, env_val in self._env.items(): + for callback in callbacks: setattr(callback, '_' + env_name, env_val) # Callback.trainer - + return callbacks + @_transfer def on_train_begin(self): pass @@ -352,6 +382,10 @@ class CallbackManager(Callback): @_transfer def on_valid_end(self, eval_result, metric_key, optimizer, is_better_eval): pass + + @_transfer + def on_validation(self): + pass @_transfer def on_epoch_end(self): @@ -366,6 +400,25 @@ class CallbackManager(Callback): pass +class DistCallbackManager(CallbackManager): + def __init__(self, env, callbacks_all=None, callbacks_master=None): + super(DistCallbackManager, self).__init__(env) + assert 'trainer' in env + is_master = env['trainer'].is_master + self.patch_callback(callbacks_master, disabled=not is_master) + self.callbacks_all = self.prepare_callbacks(callbacks_all) + self.callbacks_master = self.prepare_callbacks(callbacks_master) + self.callbacks = self.callbacks_all + self.callbacks_master + + def patch_callback(self, callbacks, disabled): + if not callbacks: + return + if not isinstance(callbacks, (list, tuple)): + callbacks = [callbacks] + for cb in callbacks: + cb._disabled = disabled + + class GradientClipCallback(Callback): """ 别名：:class:`fastNLP.GradientClipCallback` :class:`fastNLP.core.callback.GradientClipCallback` @@ -403,6 +456,9 @@ class GradientClipCallback(Callback): def on_backward_end(self): if self.step%self.update_every==0: if self.parameters is None: + if getattr(self.trainer, 'fp16', ''): + from apex import amp + self.clip_fun(amp.master_params(self.optimizer), self.clip_value) self.clip_fun(self.model.parameters(), self.clip_value) else: self.clip_fun(self.parameters, self.clip_value) @@ -448,10 +504,10 @@ class FitlogCallback(Callback): 并将验证结果写入到fitlog中。这些数据集的结果是根据dev上最好的结果报道的，即如果dev在第3个epoch取得了最佳，则 fitlog中记录的关于这些数据集的结果就是来自第三个epoch的结果。 - :param DataSet,dict(DataSet) data: 传入DataSet对象，会使用多个Trainer中的metric对数据进行验证。如果需要传入多个 + :param ~fastNLP.DataSet,Dict[~fastNLP.DataSet] data: 传入DataSet对象，会使用多个Trainer中的metric对数据进行验证。如果需要传入多个 DataSet请通过dict的方式传入，dict的key将作为对应dataset的name传递给fitlog。若tester不为None时，data需要通过 dict的方式传入。如果仅传入DataSet, 则被命名为test - :param Tester tester: Tester对象，将在on_valid_end时调用。tester中的DataSet会被称为为`test` + :param ~fastNLP.Tester tester: Tester对象，将在on_valid_end时调用。tester中的DataSet会被称为为`test` :param int log_loss_every: 多少个step记录一次loss(记录的是这几个batch的loss平均值)，如果数据集较大建议将该值设置得 大一些，不然会导致log文件巨大。默认为0, 即不要记录loss。 :param int verbose: 是否在终端打印evaluation的结果，0不打印。 @@ -479,7 +535,7 @@ class FitlogCallback(Callback): self.datasets[key] = value elif isinstance(data, DataSet): self.datasets['test'] = data - else: + elif data is not None: raise TypeError("data receives dict[DataSet] or DataSet object.") self.verbose = verbose @@ -674,7 +730,7 @@ class TensorboardCallback(Callback): .. warning:: fastNLP 已停止对此功能的维护，请等待 fastNLP 兼容 PyTorch1.1 的下一个版本。 - 或者使用和 fastNLP 高度配合的 fitlog（参见 :doc:`/user/with_fitlog` ）。 + 或者使用和 fastNLP 高度配合的 fitlog（参见 :doc:`/tutorials/tutorial_10_fitlog` ）。 """ @@ -884,3 +940,59 @@ class EarlyStopError(CallbackException): def __init__(self, msg): super(EarlyStopError, self).__init__(msg) + + +class EchoCallback(Callback): + def __init__(self, name, out=sys.stdout): + super(EchoCallback, self).__init__() + self.name = name + self.out = out + + def __getattribute__(self, item): + if item.startswith('on_'): + print('{}.{} has been called at pid: {}'.format(self.name, item, os.getpid()), + file=self.out) + return super(EchoCallback, self).__getattribute__(item) + + +class TesterCallback(Callback): + def __init__(self, data, model, metrics, metric_key=None, batch_size=16, num_workers=None): + super(TesterCallback, self).__init__() + self.tester = Tester(data, model, + metrics=metrics, batch_size=batch_size, + num_workers=num_workers, verbose=0) + # parse metric_key + # increase_better is True. It means the exp result gets better if the indicator increases. + # It is true by default. + self.increase_better = True + if metric_key is not None: + self.increase_better = False if metric_key[0] == "-" else True + self.metric_key = metric_key[1:] if metric_key[0] == "+" or metric_key[0] == "-" else metric_key + else: + self.metric_key = None + self.score = None + + def on_validation(self): + cur_score = self.tester.test() + eval_str = "Evaluation at Epoch {}/{}. Step:{}/{}. - {}".format( + self.epoch, self.n_epochs, self.step, self.n_steps, + self.tester._format_eval_results(cur_score)) + self.logger.info(eval_str) + is_better = self.compare_better(cur_score) + if is_better: + self.score = cur_score + return cur_score, is_better + + def compare_better(self, a): + if self.score is None: + return True + k = self.metric_key + is_increase = self.score[k] <= a[k] # if equal, prefer more recent results + if self.increase_better: + return is_increase + else: + return not is_increase + + def on_train_end(self): + self.logger.info('Evaluate on training ends.') + self.on_validation() diff --git a/fastNLP/core/dataset.py b/fastNLP/core/dataset.py index b7df9dec..2955eff6 100644 --- a/fastNLP/core/dataset.py +++ b/fastNLP/core/dataset.py @@ -1,7 +1,7 @@ """ :class:`~fastNLP.core.dataset.DataSet` 是fastNLP中用于承载数据的容器。可以将DataSet看做是一个表格， -每一行是一个sample (在fastNLP中被称为 :mod:`~.instance` )， -每一列是一个feature (在fastNLP中称为 :mod:`.field` )。 +每一行是一个sample (在fastNLP中被称为 :mod:`~fastNLP.core.instance` )， +每一列是一个feature (在fastNLP中称为 :mod:`~fastNLP.core.field` )。 .. csv-table:: Following is a demo layout of DataSet :header: "sentence", "words", "seq_len" @@ -13,57 +13,64 @@ 在fastNLP内部每一行是一个 :class:`~fastNLP.Instance` 对象； 每一列是一个 :class:`~fastNLP.FieldArray` 对象。 -1 DataSet的创建 - 创建DataSet主要有以下的3种方式 +---------------------------- +1.DataSet的创建 +---------------------------- -1.1 传入dict +创建DataSet主要有以下的3种方式 - Example:: +1.1 传入dict +---------------------------- - from fastNLP import DataSet - data = {'sentence':["This is the first instance .", "Second instance .", "Third instance ."], - 'words': [['this', 'is', 'the', 'first', 'instance', '.'], ['Second', 'instance', '.'], ['Third', 'instance', '.'], - 'seq_len': [6, 3, 3]} - dataset = DataSet(data) - # 传入的dict的每个key的value应该为具有相同长度的list + .. code-block:: -1.2 通过构建Instance + from fastNLP import DataSet + data = {'sentence':["This is the first instance .", "Second instance .", "Third instance ."], + 'words': [['this', 'is', 'the', 'first', 'instance', '.'], ['Second', 'instance', '.'], ['Third', 'instance', '.'], + 'seq_len': [6, 3, 3]} + dataset = DataSet(data) + # 传入的dict的每个key的value应该为具有相同长度的list - Example:: +1.2 通过 Instance 构建 +---------------------------- - from fastNLP import DataSet - from fastNLP import Instance - dataset = DataSet() - instance = Instance(sentence="This is the first instance", - words=['this', 'is', 'the', 'first', 'instance', '.'], - seq_len=6) - dataset.append(instance) - # 可以继续append更多内容，但是append的instance应该和第一个instance拥有完全相同的field + .. code-block:: -1.3 通过list(Instance) + from fastNLP import DataSet + from fastNLP import Instance + dataset = DataSet() + instance = Instance(sentence="This is the first instance", + words=['this', 'is', 'the', 'first', 'instance', '.'], + seq_len=6) + dataset.append(instance) + # 可以继续append更多内容，但是append的instance应该和第一个instance拥有完全相同的field - Example:: +1.3 通过 List[Instance] 构建 +-------------------------------------- - from fastNLP import DataSet - from fastNLP import Instance - instances = [] - instances.append(Instance(sentence="This is the first instance", - words=['this', 'is', 'the', 'first', 'instance', '.'], - seq_len=6)) - instances.append(Instance(sentence="Second instance .", - words=['Second', 'instance', '.'], - seq_len=3)) - dataset = DataSet(instances) + .. code-block:: -2 DataSet与预处理 - 常见的预处理有如下几种 + from fastNLP import DataSet + from fastNLP import Instance + instances = [] + winstances.append(Instance(sentence="This is the first instance", + ords=['this', 'is', 'the', 'first', 'instance', '.'], + seq_len=6)) + instances.append(Instance(sentence="Second instance .", + words=['Second', 'instance', '.'], + seq_len=3)) + dataset = DataSet(instances) + +-------------------------------------- +2.DataSet与预处理 +-------------------------------------- -2.1 从某个文本文件读取内容 # +常见的预处理有如下几种 - .. todo:: - 引用DataLoader +2.1 从某个文本文件读取内容 +-------------------------------------- - Example:: + .. code-block:: from fastNLP import DataSet from fastNLP import Instance @@ -78,21 +85,13 @@ sent, label = line.strip().split('\t') dataset.append(Instance(sentence=sent, label=label)) -2.2 index, 返回结果为对DataSet对象的浅拷贝 - - Example:: + .. note:: + 直接读取特定数据集的数据请参考 :doc:`/tutorials/tutorial_2_load_dataset` - import numpy as np - from fastNLP import DataSet - dataset = DataSet({'a': np.arange(10), 'b': [[_] for _ in range(10)]}) - d[0] # 使用一个下标获取一个instance - >>{'a': 0 type=int,'b': [2] type=list} # 得到一个instance - d[1:3] # 使用slice获取一个新的DataSet - >>DataSet({'a': 1 type=int, 'b': [2] type=list}, {'a': 2 type=int, 'b': [2] type=list}) +2.2 对DataSet中的内容处理 +-------------------------------------- -2.3 对DataSet中的内容处理 - - Example:: + .. code-block:: from fastNLP import DataSet data = {'sentence':["This is the first instance .", "Second instance .", "Third instance ."]} @@ -108,9 +107,10 @@ return words dataset.apply(get_words, new_field_name='words') -2.4 删除DataSet的内容 +2.3 删除DataSet的内容 +-------------------------------------- - Example:: + .. code-block:: from fastNLP import DataSet dataset = DataSet({'a': list(range(-5, 5))}) @@ -124,16 +124,18 @@ dataset.delete_field('a') -2.5 遍历DataSet的内容 +2.4 遍历DataSet的内容 +-------------------------------------- - Example:: + .. code-block:: for instance in dataset: # do something -2.6 一些其它操作 +2.5 一些其它操作 +-------------------------------------- - Example:: + .. code-block:: # 检查是否存在名为'a'的field dataset.has_field('a') # 或 ('a' in dataset) @@ -141,21 +143,25 @@ dataset.rename_field('a', 'b') # DataSet的长度 len(dataset) + +-------------------------------------- +3.DataSet与自然语言处理(NLP) +-------------------------------------- -3 DataSet与自然语言处理(NLP) - 在目前深度学习的模型中，大都依赖于随机梯度下降法(SGD)进行模型的优化。随机梯度下降需要将数据切分成一个一个的Batch， - 一个Batch进行一次前向计算(forward)与梯度后向传播(backward)。在自然语言处理的场景下，往往还需要对数据进行pad。这是 - 由于句子的长度一般是不同的，但是一次Batch中的每个field都必须是一个tensor，所以需要将所有句子都补齐到相同的长度。 +在目前深度学习的模型中，大都依赖于随机梯度下降法(SGD)进行模型的优化。随机梯度下降需要将数据切分成一个个的 batch， +一个batch进行一次前向计算(forward)与梯度后向传播(backward)。在自然语言处理的场景下，往往还需要对数据进行pad。这是 +由于句子的长度一般是不同的，但是一次batch中的每个field都必须是一个tensor，所以需要将所有句子都补齐到相同的长度。 -3.1 DataSet与Batch +3.1 DataSet与DataSetIter +-------------------------------------- - 我们先看fastNLP中如何将数据分成一个一个的Batch的例子, 这里我们使用随机生成的数据来模拟一个二分类文本分类任务， + 我们先看fastNLP中如何将数据分成一个一个的batch的例子, 这里我们使用随机生成的数据来模拟一个二分类文本分类任务， words和characters是输入，labels是文本类别 - Example:: + .. code-block:: from fastNLP import DataSet - from fastNLP import Batch + from fastNLP import DataSetIter from fastNLP import SequentialSampler from fastNLP import EngChar2DPadder @@ -175,7 +181,7 @@ d.set_target('label') d.set_input('words', 'chars') - for batch_x, batch_y in Batch(d, sampler=SequentialSampler(), batch_size=2): + for batch_x, batch_y in DataSetIter(d, sampler=SequentialSampler(), batch_size=2): print("batch_x:", batch_x) print("batch_y:", batch_y) break @@ -194,23 +200,26 @@ # [ 0, 0, 0, 0, 0]]])} # {'label': tensor([0, 0])} - 其中 :class:`~fastNLP.Batch` 是用于从DataSet中按照batch_size为大小取出batch的迭代器， - :class:`~fastNLP.SequentialSampler` 用于指示 Batch 以怎样的 + 其中 :class:`~fastNLP.DataSetIter` 是用于从DataSet中按照batch_size为大小取出batch的迭代器， + :class:`~fastNLP.SequentialSampler` 用于指示 :class:`~fastNLP.DataSetIter` 以怎样的 顺序从DataSet中取出instance以组成一个batch， - 更详细的说明请参照 :class:`~fastNLP.Batch` 和 :class:`~fastNLP.SequentialSampler` 文档。 + 更详细的说明请参照 :class:`~fastNLP.DataSetIter` 和 :class:`~fastNLP.SequentialSampler` 文档。 - 通过DataSet.set_input('words', 'chars'), fastNLP将认为'words'和'chars'这两个field都是input，并将它们都放入迭代器 - 生成的第一个dict中; DataSet.set_target('labels'), fastNLP将认为'labels'这个field是target，并将其放入到迭代器的第 + 通过 ``DataSet.set_input('words', 'chars')`` , fastNLP将认为 `words` 和 `chars` 这两个field都是input，并将它们都放入迭代器 + 生成的第一个dict中; ``DataSet.set_target('labels')`` , fastNLP将认为 `labels` 这个field是target，并将其放入到迭代器的第 二个dict中。如上例中所打印结果。分为input和target的原因是由于它们在被 :class:`~fastNLP.Trainer` 所使用时会有所差异， 详见 :class:`~fastNLP.Trainer` - 当把某个field设置为'target'或者'input'的时候(两者不是互斥的，可以同时设为input和target)，fastNLP不仅仅只是将其放 - 置到不同的dict中，而还会对被设置为input或target的field进行类型检查。类型检查的目的是为了看能否把该field转为 - pytorch的torch.LongTensor或torch.FloatTensor类型(也可以在Batch中设置输出numpy类型，参考 :class:`~fastNLP.Batch` )，如上例所示， - fastNLP已将words，chars和label转为了Tensor类型。如果field在每个instance都拥有相同的维度(不能超过两维)，且最内层 - 的元素都为相同的type(int, float, np.int*, np.float*)，则fastNLP默认将对该field进行pad。也支持全为str的field作为 - target和input，这种情况下，fastNLP默认不进行pad。另外，当某个field已经被设置为了target或者input后，之后append的 - instance对应的field必须要和前面已有的内容一致，否则会报错。 + 当把某个field设置为 `target` 或者 `input` 的时候(两者不是互斥的，可以同时设为两种)，fastNLP不仅仅只是将其放 + 置到不同的dict中，而还会对被设置为 `input` 或 `target` 的 field 进行类型检查。类型检查的目的是为了看能否把该 field 转为 + pytorch的 :class:`torch.LongTensor` 或 :class:`torch.FloatTensor` 类型 + (也可以在 :class:`~fastNLP.DataSetIter` 中设置输出numpy类型，参考 :class:`~fastNLP.DataSetIter` )。 + + 如上例所示，fastNLP已将 `words` ，`chars` 和 `label` 转为了 :class:`Tensor` 类型。 + 如果 field 在每个 `instance` 都拥有相同的维度(不能超过两维)，且最内层的元素都为相同的 type(int, float, np.int*, np.float*)， + 则fastNLP默认将对该 field 进行pad。也支持全为str的field作为target和input，这种情况下，fastNLP默认不进行pad。 + 另外，当某个 field 已经被设置为了 target 或者 input 后，之后 `append` 的 + `instance` 对应的 field 必须要和前面已有的内容一致，否则会报错。 可以查看field的dtype:: @@ -229,6 +238,7 @@ 错误:: from fastNLP import DataSet + d = DataSet({'data': [1, 'a']}) d.set_input('data') >> RuntimeError: Mixed data types in Field data: [, ] @@ -243,6 +253,7 @@ 当某个field被设置为忽略type之后，fastNLP将不对其进行pad。 3.2 DataSet与pad +-------------------------------------- 在fastNLP里，pad是与一个field绑定的。即不同的field可以使用不同的pad方式，比如在英文任务中word需要的pad和 character的pad方式往往是不同的。fastNLP是通过一个叫做 :class:`~fastNLP.Padder` 的子类来完成的。 @@ -252,7 +263,7 @@ 如果 :class:`~fastNLP.AutoPadder` 或 :class:`~fastNLP.EngChar2DPadder` 无法满足需求， 也可以自己写一个 :class:`~fastNLP.Padder` 。 - Example:: + .. code-block:: from fastNLP import DataSet from fastNLP import EngChar2DPadder @@ -417,7 +428,7 @@ class DataSet(object): """ 将一个instance对象append到DataSet后面。 - :param instance: :class:`~fastNLP.Instance` 类型。若DataSet不为空，则instance应该拥有和DataSet完全一样的field。 + :param ~fastNLP.Instance instance: 若DataSet不为空，则instance应该拥有和DataSet完全一样的field。 """ if len(self.field_arrays) == 0: @@ -443,7 +454,7 @@ class DataSet(object): 将fieldarray添加到DataSet中. :param str field_name: 新加入的field的名称 - :param fieldarray: :class:`~fastNLP.FieldArray` 类型。需要加入DataSet的field的内容 + :param ~fastNLP.core.FieldArray fieldarray: 需要加入DataSet的field的内容 :return: """ if not isinstance(fieldarray, FieldArray): @@ -459,8 +470,7 @@ class DataSet(object): :param str field_name: 新增的field的名称 :param list fields: 需要新增的field的内容 - :param None, padder: :class:`~fastNLP.Padder` 类型， - 如果为None,则不进行pad，默认使用 :class:`~fastNLP.AutoPadder` 自动判断是否需要做pad。 + :param None,~fastNLP.Padder padder: 如果为None,则不进行pad，默认使用 :class:`~fastNLP.AutoPadder` 自动判断是否需要做pad。 :param bool is_input: 新加入的field是否是input :param bool is_target: 新加入的field是否是target :param bool ignore_type: 是否忽略对新加入的field的类型检查 @@ -477,7 +487,7 @@ class DataSet(object): """ 删除第index个instance - :param int index: 需要删除的instance的index，从0开始 + :param int index: 需要删除的instance的index，序号从0开始。 """ assert isinstance(index, int), "Only integer supported." if len(self) <= index: @@ -522,7 +532,7 @@ class DataSet(object): """ 返回一个dict，key为field_name, value为对应的 :class:`~fastNLP.FieldArray` - :return: dict: 返回如上所述的字典 + :return dict: 返回如上所述的字典 """ return self.field_arrays @@ -530,7 +540,7 @@ class DataSet(object): """ 返回一个list，包含所有 field 的名字 - :return: list: 返回如上所述的列表 + :return list: 返回如上所述的列表 """ return sorted(self.field_arrays.keys()) @@ -556,7 +566,7 @@ class DataSet(object): raise KeyError("DataSet has no field named {}.".format(old_name)) return self - def set_target(self, *field_names, flag=True): + def set_target(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True): """ 将field_names的field设置为target @@ -567,11 +577,14 @@ class DataSet(object): :param str field_names: field的名称 :param bool flag: 将field_name的target状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 """ assert isinstance(flag, bool), "Only bool type supported." for name in field_names: if name in self.field_arrays: try: + self.field_arrays[name]._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self.field_arrays[name].is_target = flag except SetInputOrTargetException as e: print(f"Cannot set field:{name} as target.") @@ -579,7 +592,7 @@ class DataSet(object): else: raise KeyError("{} is not a valid field name.".format(name)) - def set_input(self, *field_names, flag=True): + def set_input(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True): """ 将field_names的field设置为input:: @@ -588,10 +601,13 @@ class DataSet(object): :param str field_names: field的名称 :param bool flag: 将field_name的input状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 """ for name in field_names: if name in self.field_arrays: try: + self.field_arrays[name]._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self.field_arrays[name].is_input = flag except SetInputOrTargetException as e: print(f"Cannot set field:{name} as input, exception happens at the {e.index} value.") @@ -624,7 +640,7 @@ class DataSet(object): dataset.set_padder('chars', padder) # 则chars这个field会使用EngChar2DPadder进行pad操作 :param str field_name: 设置field的padding方式为padder - :param None, Padder padder: 设置为None即删除padder, 即对该field不进行pad操作。 + :param None,~fastNLP.Padder padder: 设置为None即删除padder, 即对该field不进行pad操作。 """ if field_name not in self.field_arrays: raise KeyError("There is no field named {}.".format(field_name)) @@ -672,7 +688,7 @@ class DataSet(object): 2. is_target: bool, 如果为True则将名为 `new_field_name` 的field设置为target 3. ignore_type: bool, 如果为True则将名为 `new_field_name` 的field的ignore_type设置为true, 忽略其类型 - :return: list(Any), 里面的元素为func的返回值，所以list长度为DataSet的长度 + :return List[Any]: 里面的元素为func的返回值，所以list长度为DataSet的长度 """ assert len(self) != 0, "Null DataSet cannot use apply_field()." @@ -699,7 +715,7 @@ class DataSet(object): """ 将results作为加入到新的field中，field名称为new_field_name - :param list(str) results: 一般是apply*()之后的结果 + :param List[str] results: 一般是apply*()之后的结果 :param str new_field_name: 新加入的field的名称 :param dict kwargs: 用户apply*()时传入的自定义参数 :return: @@ -742,7 +758,7 @@ class DataSet(object): 3. ignore_type: bool, 如果为True则将 `new_field_name` 的field的ignore_type设置为true, 忽略其类型 - :return: list(Any), 里面的元素为func的返回值，所以list长度为DataSet的长度 + :return List[Any]: 里面的元素为func的返回值，所以list长度为DataSet的长度 """ assert len(self) != 0, "Null DataSet cannot use apply()." idx = -1 @@ -807,7 +823,7 @@ class DataSet(object): :param float ratio: 0=parse_version('1.1'): + self.model = DDP(model, device_ids=[self.local_rank], + output_device=self.local_rank, find_unused_parameters=True) + else: + self.model = DDP(model, device_ids=[self.local_rank], + output_device=self.local_rank) + + self.optimizer = optimizer + self.sampler = DistributedSampler(self.train_data) + self.data_iterator = self._get_data_iter(self.train_data) + self.n_steps = self._get_n_steps() + + # for evaluation, only run eval on master proc + if dev_data and metrics: + cb = TesterCallback( + dev_data, model, metrics, + batch_size=batch_size_per_gpu, num_workers=num_data_workers) + self.callback_manager.callbacks_master += \ + self.callback_manager.prepare_callbacks([cb]) + + # Setup logging + dist.barrier() + self.start_time = datetime.now().strftime('%m_%d_%Y-%H_%M') + if self.save_path: + self.cp_save_path = os.path.join(self.save_path, 'checkpoints', self.start_time) + else: + self.cp_save_path = None + + # use INFO in the master, WARN for others + logging.basicConfig(filename=log_path, + format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', + datefmt='%m/%d/%Y %H:%M:%S', + level=logging.INFO if self.is_master else logging.WARN) + self.logger = logging.getLogger(__name__) + self.logger.info("Setup Distributed Trainer") + self.logger.warning("Process pid: {}, rank: {}, local rank: {}, device: {}, fp16: {}".format( + os.getpid(), self.rank, self.local_rank, self.device, self.fp16 if self.fp16 else False)) + self.logger.info("Num of processes: {}".format(self.world_size)) + self.logger.info("Use device: {}".format(device)) + self.logger.info("Training with fp16: {}, optimization level: {}".format( + len(self.fp16) > 0, self.fp16 if self.fp16 else None)) + + def _get_n_steps(self): + batch_size = self.world_size * self.batch_size_per_gpu + return (len(self.train_data) // batch_size + int( + len(self.train_data) % batch_size != 0)) * int(self.drop_last == 0) * self.n_epochs + + def _get_data_iter(self, dataset): + if isinstance(dataset, DataSet): + return DataSetIter( + dataset=dataset, batch_size=self.batch_size_per_gpu, + num_workers=self.num_data_workers, sampler=self.sampler, + drop_last=self.drop_last + ) + elif isinstance(dataset, BatchIter): + return dataset + else: + raise TypeError("train_data type {} not support".format(type(dataset))) + + def _get_optimizer(self, optimizer): + if isinstance(optimizer, torch.optim.Optimizer): + return optimizer + elif isinstance(optimizer, Optimizer): + return optimizer.construct_from_pytorch(self.model.parameters()) + elif optimizer is None: + return torch.optim.Adam(self.model.parameters(), lr=4e-3) + else: + raise TypeError("optimizer can only be torch.optim.Optimizer type, not {}.".format(type(optimizer))) + + @property + def is_master(self): + return self.rank == 0 + + def train(self, on_exception='auto'): + try: + self.logger.info("###### Training epochs started ######") + self.logger.info('Total epochs: %d'% self.n_epochs) + self.logger.info('Total steps: %d'% self.n_steps) + self.logger.info('Num instances per GPU %d'% self.batch_size_per_gpu) + self.logger.info('Total batch_size: %d'% self.batch_size_per_gpu * dist.get_world_size()) + self.logger.info('Total num of samples: %d'% len(self.train_data)) + self.logger.info("Num of callbacks for all workers: {}".format( + len(self.callback_manager.callbacks_all))) + self.logger.info("Num of callbacks for master workers: {}".format( + len(self.callback_manager.callbacks_master))) + self.logger.info("Callbacks for all workers: {}".format( + [repr(cb) for cb in self.callback_manager.callbacks_all])) + self.logger.info("Callbacks for master workers: {}".format( + [repr(cb) for cb in self.callback_manager.callbacks_master])) + + start_time = time.time() + results = {} + if self.n_epochs <= 0: + self.logger.info("Training epoch is {}, nothing was done.".format(self.n_epochs)) + results['seconds'] = 0. + return results + + try: + self.callback_manager.on_train_begin() + self._train() + self.callback_manager.on_train_end() + + except BaseException as e: + self.callback_manager.on_exception(e) + if on_exception == 'auto': + if not isinstance(e, (CallbackException, KeyboardInterrupt)): + raise e + else: + self.logger.info('Catch {}, ignored.'.format(e.__class__.__name__)) + elif on_exception == 'raise': + raise e + + results['seconds'] = round(time.time() - start_time, 2) + self.logger.info("###### Train finished ######") + self.logger.info('Total train time: {} seconds.'. format(results['seconds'])) + return results + finally: + self.close() + + def _train(self): + if self.fp16: + # skip check, done in __init__() + from apex import amp + self.step = 0 + self.epoch = 0 + self.pbar = tqdm(total=self.n_steps, postfix='loss:{0:<6.5f}', + leave=False, dynamic_ncols=True, disable=not self.is_master) + pbar = self.pbar + avg_loss = 0 + data_iterator = self.data_iterator + self.model.zero_grad() + for epoch in range(1, self.n_epochs + 1): + self.epoch = epoch + pbar.set_description_str(desc="Epoch {}/{}".format(epoch, self.n_epochs)) + # early stopping + self.callback_manager.on_epoch_begin() + for batch_x, batch_y in data_iterator: + self.model.train() + self.step += 1 + _move_dict_value_to_device(batch_x, batch_y, device=self.device) + indices = data_iterator.get_batch_indices() + # negative sampling; replace unknown; re-weight batch_y + self.callback_manager.on_batch_begin(batch_x, batch_y, indices) + prediction = self._data_forward(self.model, batch_x) + + # edit prediction + self.callback_manager.on_loss_begin(batch_y, prediction) + loss = self._compute_loss(prediction, batch_y) + avg_loss += loss.item() + + # Is loss NaN or inf? requires_grad = False + self.callback_manager.on_backward_begin(loss) + + if self.fp16: + with amp.scale_loss(loss, self.optimizer) as scale_loss: + scale_loss.backward() + else: + loss.backward() + + self.callback_manager.on_backward_end() + + self._update() + self.callback_manager.on_step_end() + + if self.step % self.print_every == 0: + avg_loss = float(avg_loss) / self.print_every + print_output = "loss:{:<6.5f}".format(avg_loss) + pbar.update(self.print_every) + pbar.set_postfix_str(print_output) + avg_loss = 0 + + self.callback_manager.on_batch_end() + + if ((self.validate_every > 0 and self.step % self.validate_every == 0) or + (self.validate_every < 0 and self.step % len(data_iterator) == 0)): + self.callback_manager.on_valid_begin() + eval_res = self.callback_manager.on_validation() + eval_res = list(filter(lambda x: x is not None, eval_res)) + if len(eval_res): + eval_res, is_better = list(zip(*eval_res)) + else: + eval_res, is_better = None, None + self.callback_manager.on_valid_end( + eval_res, self.metric_key, self.optimizer, is_better) + dist.barrier() + + if self.cp_save_path and \ + self.save_every > 0 and \ + self.step % self.save_every == 0: + self.save_check_point() + + # ================= mini-batch end ==================== # + if self.save_every < 0 and self.cp_save_path: + self.save_check_point() + # lr decay; early stopping + self.callback_manager.on_epoch_end() + # =============== epochs end =================== # + pbar.close() + self.pbar = None + # ============ tqdm end ============== # + + def _update(self): + """Perform weight update on a model. + + """ + if self.step % self.update_every == 0: + self.optimizer.step() + self.model.zero_grad() + + def _data_forward(self, network, x): + x = _build_args(self._forward_func, **x) + y = network(**x) + if not isinstance(y, dict): + raise TypeError( + f"The return value of {_get_func_signature(self._forward_func)} should be dict, got {type(y)}.") + return y + + def _compute_loss(self, predict, truth): + """Compute loss given prediction and ground truth. + + :param predict: prediction dict, produced by model.forward + :param truth: ground truth dict, produced by batch_y + :return: a scalar + """ + loss = self.losser(predict, truth) + if self.update_every > 1: + loss = loss / self.update_every + return loss.mean() + + def save_check_point(self, only_params=False): + # only master save models + if self.is_master: + os.makedirs(self.cp_save_path, exist_ok=True) + path = os.path.join(self.cp_save_path, 'checkpoint-{}.bin'.format(self.step)) + self.logger.info("Save checkpoint to {}".format(path)) + model_to_save = self.model.module + if only_params: + model_to_save = model_to_save.state_dict() + torch.save(model_to_save, path) + + def close(self): + dist.destroy_process_group() diff --git a/fastNLP/core/field.py b/fastNLP/core/field.py index 65eb0194..d7d3bb8b 100644 --- a/fastNLP/core/field.py +++ b/fastNLP/core/field.py @@ -23,7 +23,8 @@ class AppendToTargetOrInputException(Exception): self.field_name = field_name # 标示当前field的名称 class FieldArray: - def __init__(self, name, content, is_target=False, is_input=False, padder=None, ignore_type=False): + def __init__(self, name, content, is_target=False, is_input=False, padder=None, ignore_type=False, + use_1st_ins_infer_dim_type=True): if len(content)==0: raise RuntimeError("Empty fieldarray is not allowed.") _content = content @@ -38,6 +39,7 @@ class FieldArray: # 根据input的情况设置input，target等 self._cell_ndim = None # 多少维度 self.dtype = None # 最内层的element都是什么类型的 + self._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self._is_input = False self._is_target = False @@ -77,7 +79,7 @@ class FieldArray: if value is True and \ self._is_target is False and \ self._ignore_type is False: - self._check_dtype_and_ndim() + self._check_dtype_and_ndim(only_check_1st_ins_dim_type=self._use_1st_ins_infer_dim_type) if value is False and self._is_target is False: self.dtype = None self._cell_ndim = None @@ -95,32 +97,34 @@ class FieldArray: if value is True and \ self._is_input is False and \ self._ignore_type is False: - self._check_dtype_and_ndim() + self._check_dtype_and_ndim(only_check_1st_ins_dim_type=self._use_1st_ins_infer_dim_type) if value is False and self._is_input is False: self.dtype = None self._cell_ndim = None self._is_target = value - def _check_dtype_and_ndim(self): + def _check_dtype_and_ndim(self, only_check_1st_ins_dim_type=True): """ 检查当前content所有的element是否是同一个类型，且是否每个元素具有相同的维度。通过的话，设置_cell_ndim与_ele_type属性；没有 通过将直接报错. + :param bool only_check_1st_ins_dim_type: 是否只检查第一个元素的type和dim :return: """ cell_0 = self.content[0] index = 0 try: type_0, dim_0 = _get_ele_type_and_dim(cell_0) - for cell in self.content[1:]: - index += 1 - type_i, dim_i = _get_ele_type_and_dim(cell) - if type_i!=type_0: - raise SetInputOrTargetException("Type:{} in index {} is different from the first element with type:{}." - ".".format(type_i, index, type_0)) - if dim_0!=dim_i: - raise SetInputOrTargetException("Dimension:{} in index {} is different from the first element with " - "dimension:{}.".format(dim_i, index, dim_0)) + if not only_check_1st_ins_dim_type: + for cell in self.content[1:]: + index += 1 + type_i, dim_i = _get_ele_type_and_dim(cell) + if type_i!=type_0: + raise SetInputOrTargetException("Type:{} in index {} is different from the first element with type:{}." + ".".format(type_i, index, type_0)) + if dim_0!=dim_i: + raise SetInputOrTargetException("Dimension:{} in index {} is different from the first element with " + "dimension:{}.".format(dim_i, index, dim_0)) self._cell_ndim = dim_0 self.dtype = type_0 except SetInputOrTargetException as e: @@ -132,7 +136,7 @@ class FieldArray: :param val: 把该val append到fieldarray。 :return: """ - if (self._is_target or self._is_input) and self._ignore_type is False: + if (self._is_target or self._is_input) and self._ignore_type is False and not self._use_1st_ins_infer_dim_type: type_, dim_ = _get_ele_type_and_dim(val) if self.dtype!=type_: raise AppendToTargetOrInputException(f"Value(type:{type_}) are of different types with " @@ -144,6 +148,14 @@ class FieldArray: else: self.content.append(val) + def pop(self, index): + """ + 删除该field中index处的元素 + :param int index: 从0开始的数据下标。 + :return: + """ + self.content.pop(index) + def __getitem__(self, indices): return self.get(indices, pad=False) @@ -448,9 +460,10 @@ class Padder: 用于对batch进行padding操作。传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前deepcopy一份。 .. py:function:: __call__(self, contents, field_name, field_ele_dtype): + 传入的是List内容。假设有以下的DataSet。 - :param list(Any) contents: 传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前 + :param List[Any] contents: 传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前 deepcopy一份。 :param str, field_name: field的名称。 :param np.int64,np.float64,np.str,None, field_ele_dtype: 该field的内层元素的类型。如果该field的ignore_type为True，该这个值为None。 @@ -469,7 +482,7 @@ class Padder: """ 传入的是List内容。假设有以下的DataSet。 - :param list(Any) contents: 传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前 + :param List[Any] contents: 传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前 deepcopy一份。 :param str, field_name: field的名称。 :param np.int64,np.float64,np.str,None, field_ele_dtype: 该field的内层元素的类型。如果该field的ignore_type为True， diff --git a/fastNLP/core/losses.py b/fastNLP/core/losses.py index 14aacef0..21c024f0 100644 --- a/fastNLP/core/losses.py +++ b/fastNLP/core/losses.py @@ -208,7 +208,7 @@ class CrossEntropyLoss(LossBase): :param seq_len: 句子的长度, 长度之外的token不会计算loss。。 :param padding_idx: padding的index，在计算loss时将忽略target中标号为padding_idx的内容, 可以通过该值代替 传入seq_len. - :param str reduction: 支持'mean'，'sum'和'none'. + :param str reduction: 支持 `mean` ，`sum` 和 `none` . Example:: @@ -225,7 +225,7 @@ class CrossEntropyLoss(LossBase): def get_loss(self, pred, target, seq_len=None): if pred.dim() > 2: - if pred.size(1) != target.size(1): + if pred.size(1) != target.size(1): # 有可能顺序替换了 pred = pred.transpose(1, 2) pred = pred.reshape(-1, pred.size(-1)) target = target.reshape(-1) @@ -265,9 +265,9 @@ class BCELoss(LossBase): 二分类交叉熵损失函数 - :param pred: 参数映射表中`pred`的映射关系，None表示映射关系为`pred`->`pred` - :param target: 参数映射表中`target`的映射关系，None表示映射关系为`target`->`target` - :param str reduction: 支持'mean'，'sum'和'none'. + :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` + :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` + :param str reduction: 支持 `mean` ，`sum` 和 `none` . """ def __init__(self, pred=None, target=None, reduction='mean'): @@ -286,11 +286,11 @@ class NLLLoss(LossBase): 负对数似然损失函数 - :param pred: 参数映射表中`pred`的映射关系，None表示映射关系为`pred`->`pred` - :param target: 参数映射表中`target`的映射关系，None表示映射关系为`target`->`target` + :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` + :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` :param ignore_idx: ignore的index，在计算loss时将忽略target中标号为ignore_idx的内容, 可以通过该值代替 传入seq_len. - :param str reduction: 支持'mean'，'sum'和'none'. + :param str reduction: 支持 `mean` ，`sum` 和 `none` . """ def __init__(self, pred=None, target=None, ignore_idx=-100, reduction='mean'): diff --git a/fastNLP/core/metrics.py b/fastNLP/core/metrics.py index f75b6c90..94f50253 100644 --- a/fastNLP/core/metrics.py +++ b/fastNLP/core/metrics.py @@ -27,14 +27,14 @@ from abc import abstractmethod class MetricBase(object): """ - 所有metrics的基类,，所有的传入到Trainer, Tester的Metric需要继承自该对象，需要覆盖写入evaluate(), get_metric()方法。 + 所有metrics的基类,所有的传入到Trainer, Tester的Metric需要继承自该对象，需要覆盖写入evaluate(), get_metric()方法。 evaluate(xxx)中传入的是一个batch的数据。 get_metric(xxx)当所有数据处理完毕，调用该方法得到最终的metric值 以分类问题中，Accuracy计算为例 - 假设model的forward返回dict中包含'pred'这个key, 并且该key需要用于Accuracy:: + 假设model的forward返回dict中包含 `pred` 这个key, 并且该key需要用于Accuracy:: class Model(nn.Module): def __init__(xxx): @@ -43,7 +43,7 @@ class MetricBase(object): # do something return {'pred': pred, 'other_keys':xxx} # pred's shape: batch_size x num_classes - 假设dataset中'label'这个field是需要预测的值，并且该field被设置为了target + 假设dataset中 `label` 这个field是需要预测的值，并且该field被设置为了target 对应的AccMetric可以按如下的定义, version1, 只使用这一次:: class AccMetric(MetricBase): @@ -478,7 +478,7 @@ class SpanFPreRecMetric(MetricBase): 别名：:class:`fastNLP.SpanFPreRecMetric` :class:`fastNLP.core.metrics.SpanFPreRecMetric` 在序列标注问题中，以span的方式计算F, pre, rec. - 比如中文Part of speech中，会以character的方式进行标注，句子'中国在亚洲'对应的POS可能为(以BMES为例) + 比如中文Part of speech中，会以character的方式进行标注，句子 `中国在亚洲` 对应的POS可能为(以BMES为例) ['B-NN', 'E-NN', 'S-DET', 'B-NN', 'E-NN']。该metric就是为类似情况下的F1计算。 最后得到的metric结果为:: @@ -502,15 +502,15 @@ class SpanFPreRecMetric(MetricBase): :param tag_vocab: 标签的 :class:`~fastNLP.Vocabulary` 。支持的标签为"B"(没有label)；或"B-xxx"(xxx为某种label，比如POS中的NN)， 在解码时，会将相同xxx的认为是同一个label，比如['B-NN', 'E-NN']会被合并为一个'NN'. - :param str pred: 用该key在evaluate()时从传入dict中取出prediction数据。 为None，则使用'pred'取数据 - :param str target: 用该key在evaluate()时从传入dict中取出target数据。 为None，则使用'target'取数据 - :param str seq_len: 用该key在evaluate()时从传入dict中取出sequence length数据。为None，则使用'seq_len'取数据。 + :param str pred: 用该key在evaluate()时从传入dict中取出prediction数据。 为None，则使用 `pred` 取数据 + :param str target: 用该key在evaluate()时从传入dict中取出target数据。 为None，则使用 `target` 取数据 + :param str seq_len: 用该key在evaluate()时从传入dict中取出sequence length数据。为None，则使用 `seq_len` 取数据。 :param str encoding_type: 目前支持bio, bmes, bmeso, bioes :param list ignore_labels: str 组成的list. 这个list中的class不会被用于计算。例如在POS tagging时传入['NN']，则不会计算'NN'这 个label :param bool only_gross: 是否只计算总的f1, precision, recall的值；如果为False，不仅返回总的f1, pre, rec, 还会返回每个 label的f1, pre, rec - :param str f_type: 'micro'或'macro'. 'micro':通过先计算总体的TP，FN和FP的数量，再计算f, precision, recall; 'macro': + :param str f_type: `micro` 或 `macro` . `micro` :通过先计算总体的TP，FN和FP的数量，再计算f, precision, recall; `macro` : 分布计算每个类别的f, precision, recall，然后做平均（各类别f的权重相同） :param float beta: f_beta分数， :math:`f_{beta} = \frac{(1 + {beta}^{2})*(pre*rec)}{({beta}^{2}*pre + rec)}` . 常用为beta=0.5, 1, 2. 若为0.5则精确率的权重高于召回率；若为1，则两者平等；若为2，则召回率权重高于精确率。 @@ -814,8 +814,8 @@ class ExtractiveQAMetric(MetricBase): if not self.right_open: e += 1 te += 1 - if ts == 0 and te == int(not self.right_open): - if s == 0 and e == int(not self.right_open): + if ts == 0 and te == 1: + if s == 0 and e == 1: self.no_ans_correct += 1 self.no2no += 1 else: diff --git a/fastNLP/core/optimizer.py b/fastNLP/core/optimizer.py index 1fe035bf..e95047b4 100644 --- a/fastNLP/core/optimizer.py +++ b/fastNLP/core/optimizer.py @@ -5,7 +5,8 @@ optimizer 模块定义了 fastNLP 中所需的各种优化器，一般做为 :cl __all__ = [ "Optimizer", "SGD", - "Adam" + "Adam", + "AdamW" ] import torch @@ -48,7 +49,7 @@ class NullOptimizer(Optimizer): super().__init__(None) def construct_from_pytorch(self, model_params): - pass + return self def __getattr__(self, item): def pass_func(*args, **kwargs): @@ -103,21 +104,28 @@ class Adam(Optimizer): class AdamW(TorchOptimizer): - r"""对AdamW的实现，该实现应该会在pytorch更高版本中出现，https://github.com/pytorch/pytorch/pull/21250。这里提前加入 + r""" + 别名：:class:`fastNLP.AdamW` :class:`fastNLP.core.optimizer.AdamW` + + 对AdamW的实现，该实现应该会在pytorch更高版本中出现，https://github.com/pytorch/pytorch/pull/21250。这里提前加入 + + .. todo:: + 翻译成中文 + The original Adam algorithm was proposed in `Adam: A Method for Stochastic Optimization`_. The AdamW variant was proposed in `Decoupled Weight Decay Regularization`_. - Arguments: - params (iterable): iterable of parameters to optimize or dicts defining - parameter groups - lr (float, optional): learning rate (default: 1e-3) - betas (Tuple[float, float], optional): coefficients used for computing - running averages of gradient and its square (default: (0.9, 0.99)) - eps (float, optional): term added to the denominator to improve - numerical stability (default: 1e-8) - weight_decay (float, optional): weight decay coefficient (default: 1e-2) - amsgrad (boolean, optional): whether to use the AMSGrad variant of this - algorithm from the paper `On the Convergence of Adam and Beyond`_ - (default: False) + + :param params (iterable): iterable of parameters to optimize or dicts defining + parameter groups + :param lr (float, optional): learning rate (default: 1e-3) + :param betas (Tuple[float, float], optional): coefficients used for computing + running averages of gradient and its square (default: (0.9, 0.99)) + :param eps (float, optional): term added to the denominator to improve + numerical stability (default: 1e-8) + :param weight_decay (float, optional): weight decay coefficient (default: 1e-2) + algorithm from the paper `On the Convergence of Adam and Beyond`_ + (default: False) + .. _Adam\: A Method for Stochastic Optimization: https://arxiv.org/abs/1412.6980 .. _Decoupled Weight Decay Regularization: @@ -147,9 +155,9 @@ class AdamW(TorchOptimizer): def step(self, closure=None): """Performs a single optimization step. - Arguments: - closure (callable, optional): A closure that reevaluates the model - and returns the loss. + + :param closure: (callable, optional) A closure that reevaluates the model + and returns the loss. """ loss = None if closure is not None: diff --git a/fastNLP/core/sampler.py b/fastNLP/core/sampler.py index c5784f59..9ca04fa0 100644 --- a/fastNLP/core/sampler.py +++ b/fastNLP/core/sampler.py @@ -25,9 +25,9 @@ class Sampler(object): def __call__(self, data_set): """ - :param DataSet data_set: `DataSet` 对象, 需要Sample的数据 - :return result: list(int) 其中元素的下标序列, ``data_set`` 中元素会按 ``result`` 中顺序取出 - """ + :param DataSet data_set: `DataSet` 对象, 需要Sample的数据 + :return result: list(int) 其中元素的下标序列, ``data_set`` 中元素会按 ``result`` 中顺序取出 + """ raise NotImplementedError @@ -62,16 +62,27 @@ class BucketSampler(Sampler): 带Bucket的 `Random Sampler`. 可以随机地取出长度相似的元素 :param int num_buckets: bucket的数量 - :param int batch_size: batch的大小 + :param int batch_size: batch的大小. 默认为None，Trainer在调用BucketSampler时，会将该值正确设置，如果是非Trainer场景使用，需 + 要显示传递该值 :param str seq_len_field_name: 对应序列长度的 `field` 的名字 """ - def __init__(self, num_buckets=10, batch_size=32, seq_len_field_name='seq_len'): + def __init__(self, num_buckets=10, batch_size=None, seq_len_field_name='seq_len'): self.num_buckets = num_buckets self.batch_size = batch_size self.seq_len_field_name = seq_len_field_name - + + def set_batch_size(self, batch_size): + """ + + :param int batch_size: 每个batch的大小 + :return: + """ + self.batch_size = batch_size + def __call__(self, data_set): + if self.batch_size is None: + raise RuntimeError("batch_size is None.") seq_lens = data_set.get_all_fields()[self.seq_len_field_name].content total_sample_num = len(seq_lens) diff --git a/fastNLP/core/tester.py b/fastNLP/core/tester.py index 7048d0ae..691bf2ae 100644 --- a/fastNLP/core/tester.py +++ b/fastNLP/core/tester.py @@ -1,7 +1,7 @@ """ tester模块实现了 fastNLP 所需的Tester类，能在提供数据、模型以及metric的情况下进行性能测试。 -Example:: +.. code-block:: import numpy as np import torch @@ -32,9 +32,16 @@ Tester在验证进行之前会调用model.eval()提示当前进入了evaluation """ +import time + import torch import torch.nn as nn +try: + from tqdm.auto import tqdm +except: + from .utils import _pseudo_tqdm as tqdm + from .batch import BatchIter, DataSetIter from .dataset import DataSet from .metrics import _prepare_metrics @@ -47,6 +54,7 @@ from .utils import _get_func_signature from .utils import _get_model_device from .utils import _move_model_to_device from ._parallel_utils import _data_parallel_wrapper +from ._parallel_utils import _model_contains_inner_module from functools import partial __all__ = [ @@ -60,15 +68,14 @@ class Tester(object): Tester是在提供数据，模型以及metric的情况下进行性能测试的类。需要传入模型，数据以及metric进行验证。 - :param data: 需要测试的数据集， :class:`~fastNLP.DataSet` 类型 + :param ~fastNLP.DataSet data: 需要测试的数据集 :param torch.nn.module model: 使用的模型 - :param metrics: :class:`~fastNLP.core.metrics.MetricBase` 或者一个列表的 :class:`~fastNLP.core.metrics.MetricBase` + :param ~fastNLP.core.metrics.MetricBase,List[~fastNLP.core.metrics.MetricBase] metrics: 测试时使用的metrics :param int batch_size: evaluation时使用的batch_size有多大。 :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 的计算位置进行管理。支持以下的输入: - 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中, 可见的第一个GPU中, - 可见的第二个GPU中; + 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中,可见的第一个GPU中,可见的第二个GPU中; 2. torch.device：将模型装载到torch.device上。 @@ -80,13 +87,12 @@ class Tester(object): 如果模型是通过predict()进行预测的话，那么将不能使用多卡(DataParallel)进行验证，只会使用第一张卡上的模型。 :param int verbose: 如果为0不输出任何信息; 如果为1，打印出验证结果。 + :param bool use_tqdm: 是否使用tqdm来显示测试进度; 如果为False，则不会显示任何内容。 """ - def __init__(self, data, model, metrics, batch_size=16, num_workers=0, device=None, verbose=1): + def __init__(self, data, model, metrics, batch_size=16, num_workers=0, device=None, verbose=1, use_tqdm=True): super(Tester, self).__init__() - - if not isinstance(data, DataSet): - raise TypeError(f"The type of data must be `fastNLP.DataSet`, got `{type(data)}`.") + if not isinstance(model, nn.Module): raise TypeError(f"The type of model must be `torch.nn.Module`, got `{type(model)}`.") @@ -96,6 +102,7 @@ class Tester(object): self._model = _move_model_to_device(model, device=device) self.batch_size = batch_size self.verbose = verbose + self.use_tqdm = use_tqdm if isinstance(data, DataSet): self.data_iterator = DataSetIter( @@ -107,19 +114,22 @@ class Tester(object): # check predict if (hasattr(self._model, 'predict') and callable(self._model.predict)) or \ - (isinstance(self._model, nn.DataParallel) and hasattr(self._model.module, 'predict') and - callable(self._model.module.predict)): + (_model_contains_inner_module(self._model) and hasattr(self._model.module, 'predict') and + callable(self._model.module.predict)): if isinstance(self._model, nn.DataParallel): self._predict_func_wrapper = partial(_data_parallel_wrapper('predict', self._model.device_ids, self._model.output_device), network=self._model.module) + self._predict_func = self._model.module.predict # 用于匹配参数 + elif isinstance(self._model, nn.parallel.DistributedDataParallel): self._predict_func = self._model.module.predict + self._predict_func_wrapper = self._model.module.predict # 用于调用 else: self._predict_func = self._model.predict self._predict_func_wrapper = self._model.predict else: - if isinstance(self._model, nn.DataParallel): + if _model_contains_inner_module(model): self._predict_func_wrapper = self._model.forward self._predict_func = self._model.module.forward else: @@ -140,21 +150,39 @@ class Tester(object): eval_results = {} try: with torch.no_grad(): - for batch_x, batch_y in data_iterator: - _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) - pred_dict = self._data_forward(self._predict_func, batch_x) - if not isinstance(pred_dict, dict): - raise TypeError(f"The return value of {_get_func_signature(self._predict_func)} " - f"must be `dict`, got {type(pred_dict)}.") + if not self.use_tqdm: + from .utils import _pseudo_tqdm as inner_tqdm + else: + inner_tqdm = tqdm + with inner_tqdm(total=len(data_iterator), leave=False, dynamic_ncols=True) as pbar: + pbar.set_description_str(desc="Test") + + start_time = time.time() + + for batch_x, batch_y in data_iterator: + _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) + pred_dict = self._data_forward(self._predict_func, batch_x) + if not isinstance(pred_dict, dict): + raise TypeError(f"The return value of {_get_func_signature(self._predict_func)} " + f"must be `dict`, got {type(pred_dict)}.") + for metric in self.metrics: + metric(pred_dict, batch_y) + + if self.use_tqdm: + pbar.update() + for metric in self.metrics: - metric(pred_dict, batch_y) - for metric in self.metrics: - eval_result = metric.get_metric() - if not isinstance(eval_result, dict): - raise TypeError(f"The return value of {_get_func_signature(metric.get_metric)} must be " - f"`dict`, got {type(eval_result)}") - metric_name = metric.__class__.__name__ - eval_results[metric_name] = eval_result + eval_result = metric.get_metric() + if not isinstance(eval_result, dict): + raise TypeError(f"The return value of {_get_func_signature(metric.get_metric)} must be " + f"`dict`, got {type(eval_result)}") + metric_name = metric.__class__.__name__ + eval_results[metric_name] = eval_result + + end_time = time.time() + test_str = f'Evaluate data in {round(end_time - start_time, 2)} seconds!' + pbar.write(test_str) + pbar.close() except _CheckError as e: prev_func_signature = _get_func_signature(self._predict_func) _check_loss_evaluate(prev_func_signature=prev_func_signature, func_signature=e.func_signature, diff --git a/fastNLP/core/trainer.py b/fastNLP/core/trainer.py index eabda99c..a85b7fee 100644 --- a/fastNLP/core/trainer.py +++ b/fastNLP/core/trainer.py @@ -11,288 +11,310 @@ Trainer在fastNLP中用于组织单任务的训练过程，可以避免用户在 (5) 保存获得更好验证性能的模型。 -1 Trainer的基本使用 - 下面的例子是使用神经网络来进行预测一个序列中是否有偶数个1。 - - Example:: - - import numpy as np - from torch import nn - import torch - import torch.nn.functional as F - from torch.optim import SGD - - from fastNLP import DataSet - from fastNLP import Trainer - from fastNLP import CrossEntropyLoss - from fastNLP import AccuracyMetric - from fastNLP.modules.decoder import MLP - - # 模型 - class Model(nn.Module): - def __init__(self, input_num): - super().__init__() - self.fcs = MLP([input_num, 40, 40, 2], 'relu') - - def forward(self, x): - x = self.fcs(x) - return {'pred': x} - model = Model(10) - - # 生成数据 - def generate_psedo_dataset(num_samples): - dataset = DataSet() - data = np.random.randint(2, size=(num_samples, 10)) - label = np.sum(data, axis=1)%2 - dataset = DataSet({'x':data.astype(float), 'label': label}) - dataset.set_input('x') - dataset.set_target('label') - return dataset - tr_dataset = generate_psedo_dataset(1000) - dev_data = generate_psedo_dataset(100) - - # 训练 - trainer = Trainer(tr_dataset, model, loss=CrossEntropyLoss(target='label'), - optimizer=SGD(model.parameters(), lr=0.1),n_epochs=1000, - dev_data = dev_data, metrics=AccuracyMetric(target='label')) - trainer.train() - - 由上面的例子可以看出通过使用Trainer，可以使得训练部分的代码大幅减少。 - 使用Trainer需要满足以下几个条件: + +---------------------------- +1. Trainer的基本使用 +---------------------------- + +下面的例子是使用神经网络来进行预测一个序列中是否有偶数个1。 + +.. code-block:: python + + import numpy as np + from torch import nn + import torch + import torch.nn.functional as F + from torch.optim import SGD + + from fastNLP import DataSet + from fastNLP import Trainer + from fastNLP import CrossEntropyLoss + from fastNLP import AccuracyMetric + from fastNLP.modules.decoder import MLP + + # 模型 + class Model(nn.Module): + def __init__(self, input_num): + super().__init__() + self.fcs = MLP([input_num, 40, 40, 2], 'relu') + + def forward(self, x): + x = self.fcs(x) + return {'pred': x} + model = Model(10) + + # 生成数据 + def generate_psedo_dataset(num_samples): + dataset = DataSet() + data = np.random.randint(2, size=(num_samples, 10)) + label = np.sum(data, axis=1)%2 + dataset = DataSet({'x':data.astype(float), 'label': label}) + dataset.set_input('x') + dataset.set_target('label') + return dataset + tr_dataset = generate_psedo_dataset(1000) + dev_data = generate_psedo_dataset(100) + + # 训练 + trainer = Trainer(tr_dataset, model, loss=CrossEntropyLoss(target='label'), + optimizer=SGD(model.parameters(), lr=0.1),n_epochs=1000, + dev_data = dev_data, metrics=AccuracyMetric(target='label')) + trainer.train() + +由上面的例子可以看出通过使用Trainer，可以使得训练部分的代码大幅减少。 +使用Trainer需要满足以下几个条件: 1.1 模型 - 1 模型的forward()的参数名需要与DataSet中的名字对应。实际上fastNLP在将DataSet中的数据传递给模型forward()时，是 - 通过匹配名称实现的。所以上例中，如果Model的forward函数修改为forward(self, data), 则DataSet中的'x'这个field就应该 - 改名为'data'。 +---------------------------- + +1 模型的forward()的参数名需要与DataSet中的名字对应。实际上fastNLP在将DataSet中的数据传递给模型forward()时，是 +通过匹配名称实现的。所以上例中，如果Model的forward函数修改为forward(self, data), 则DataSet中的'x'这个field就应该 +改名为'data'。 - 2 传递给forward()的参数是DataSet中被设置为input的那些field。但如果forward()中没有对应的参数，则不会将数据传递 - 给forward()。例如，DataSet中'x1', 'x2'都是input，但是模型的函数为forward(self, x1), 那么'x2'不会传递给forward()。 +2 传递给forward()的参数是DataSet中被设置为input的那些field。但如果forward()中没有对应的参数，则不会将数据传递 +给forward()。例如，DataSet中'x1', 'x2'都是input，但是模型的函数为forward(self, x1), 那么'x2'不会传递给forward()。 - 3 模型的forward()返回值需要为一个dict。 +3 模型的forward()返回值需要为一个dict。 1.2 Loss - fastNLP中的为了不限制forward函数的返回内容数量(比如一些复杂任务需要返回多个内容，如Dependency Parsing， - :mod:`Loss` 与 :mod:`Metric` 都使用了通过名称来匹配相应内容的策略。如上面的例子中 +---------------------------- - Example:: +fastNLP中的为了不限制forward函数的返回内容数量(比如一些复杂任务需要返回多个内容，如Dependency Parsing， +:mod:`Loss` 与 :mod:`Metric` 都使用了通过名称来匹配相应内容的策略。如上面的例子中 - trainer = Trainer(tr_dataset, model, loss=CrossEntropyLoss(target='label'), - optimizer=SGD(model.parameters(), lr=0.1),n_epochs=1000, - dev_data = dev_data, metrics=AccuracyMetric(target='label')) +.. code-block:: python - loss被设置为了 :class:`~fastNLP.CrossEntropyLoss` , 但在初始化的时候传入了target='label'这个参数， - :class:`~fastNLP.CrossEntropyLoss` 的初始化参数为(pred=None, target=None, padding_idx=-100)。 - - 这里的两个参数分别为计算CrossEntropy时需要使用到的模型的预测值与真实值。 - 其中 `pred` 一般来自于模型forward()的返回结果，`target` 一般是来自于DataSet中被设置为target的field。 - 由于每个人对真实值或者model的返回值取名并不一样，所以fastNLP的 :mod:`Loss` 提供一种类似于映射的机制来匹配对应的值， - 比如这里 :class:`~fastNLP.CrossEntropyLoss` 将尝试找到名为'label'的内容来作为真实值得到loss； - 而pred=None, 则 :class:`~fastNLP.CrossEntropyLoss` 使用'pred'作为名称匹配预测值， - 正好forward的返回值也叫pred，所以这里不需要申明pred。 - - 尽管fastNLP使用了映射机制来使得loss的计算变得比较灵活，但有些情况下loss必须在模型中进行计算，比如使用了CRF的模型。 - fastNLP中提供了 :class:`~fastNLP.LossInForward` 这个loss。 - 这个loss的原理是直接在forward()的返回结果中找到loss_key(默认寻找'loss')指定的那个tensor，并使用它作为loss。 - 如果Trainer初始化没有提供loss则默认使用 :class:`~fastNLP.LossInForward` 。 - - .. todo:: - 补充一个例子 详细例子可以参照 + trainer = Trainer(tr_dataset, model, loss=CrossEntropyLoss(target='label'), + optimizer=SGD(model.parameters(), lr=0.1),n_epochs=1000, + dev_data = dev_data, metrics=AccuracyMetric(target='label')) + +loss被设置为了 :class:`~fastNLP.CrossEntropyLoss` , 但在初始化的时候传入了target='label'这个参数， +:class:`~fastNLP.CrossEntropyLoss` 的初始化参数为(pred=None, target=None, padding_idx=-100)。 + +这里的两个参数分别为计算CrossEntropy时需要使用到的模型的预测值与真实值。 +其中 `pred` 一般来自于模型forward()的返回结果，`target` 一般是来自于DataSet中被设置为target的field。 +由于每个人对真实值或者model的返回值取名并不一样，所以fastNLP的 :mod:`Loss` 提供一种类似于映射的机制来匹配对应的值， +比如这里 :class:`~fastNLP.CrossEntropyLoss` 将尝试找到名为'label'的内容来作为真实值得到loss； +而pred=None, 则 :class:`~fastNLP.CrossEntropyLoss` 使用'pred'作为名称匹配预测值， +正好forward的返回值也叫pred，所以这里不需要申明pred。 + +尽管fastNLP使用了映射机制来使得loss的计算变得比较灵活，但有些情况下loss必须在模型中进行计算，比如使用了CRF的模型。 +fastNLP中提供了 :class:`~fastNLP.LossInForward` 这个loss。 +这个loss的原理是直接在forward()的返回结果中找到loss_key(默认寻找'loss')指定的那个tensor，并使用它作为loss。 +如果Trainer初始化没有提供loss则默认使用 :class:`~fastNLP.LossInForward` 。 + +.. todo:: + 补充一个例子 详细例子可以参照 1.3 Metric - :mod:`Metric` 使用了与上述Loss一样的策略，即使用名称进行匹配。 - AccuracyMetric(target='label')的情况与CrossEntropyLoss 是同理的。 - - 在进行验证时，可能用到的计算与forward()中不太一致，没有办法直接从forward()的结果中得到预测值，这时模型可以提供一个predict()方法， - 如果提供的模型具有predict方法，则在模型验证时将调用predict()方法获取预测结果， - 传入到predict()的参数也是从DataSet中被设置为input的field中选择出来的; - 与forward()一样，返回值需要为一个dict。 +---------------------------- + +:mod:`Metric` 使用了与上述Loss一样的策略，即使用名称进行匹配。 +AccuracyMetric(target='label')的情况与CrossEntropyLoss 是同理的。 + +在进行验证时，可能用到的计算与forward()中不太一致，没有办法直接从forward()的结果中得到预测值，这时模型可以提供一个predict()方法， +如果提供的模型具有predict方法，则在模型验证时将调用predict()方法获取预测结果， +传入到predict()的参数也是从DataSet中被设置为input的field中选择出来的; +与forward()一样，返回值需要为一个dict。 + +.. todo:: + 补充一个例子 具体例子可以参考 - .. todo:: - 补充一个例子 具体例子可以参考 +---------------------------- +2. Trainer的代码检查 +---------------------------- -2 Trainer的代码检查 - 由于在fastNLP中采取了映射的机制，所以难免可能存在对应出错的情况。Trainer提供一种映射检查机制，可以通过check_code_level来进行控制 - 比如下面的例子中，由于各种原因产生的报错 +由于在fastNLP中采取了映射的机制，所以难免可能存在对应出错的情况。Trainer提供一种映射检查机制，可以通过check_code_level来进行控制 +比如下面的例子中，由于各种原因产生的报错 Example2.1 - :: - - import numpy as np - from torch import nn - import torch - from torch.optim import SGD - from fastNLP import Trainer - from fastNLP import DataSet - - class Model(nn.Module): - def __init__(self): - super().__init__() - self.fc = nn.Linear(1, 1) - def forward(self, x, b): - loss = torch.mean((self.fc(x)-b)**2) - return {'loss': loss} - model = Model() - - dataset = DataSet({'a': np.arange(10), 'b':np.arange(10)*2}) - dataset.set_input('a', 'b') - - trainer = Trainer(dataset, model, loss=None, optimizer=SGD(model.parameters(), lr=0.001)) - - trainer = Trainer(dataset, model, SGD(model.parameters())) - # 会报以下的错误 - # input fields after batch(if batch size is 2): - # a: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) - # b: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) - # There is no target field. - # .... - # NameError: - # Problems occurred when calling Model.forward(self, x, b) - # missing param: ['x'] - # unused field: ['a'] - # Suggestion: You need to provide ['x'] in DataSet and set it as input. - - 这里就是由于在Trainer初始化的时候，fastNLP会尝试使用一个batch_size=2的batch去运行一遍forward()以及backward()。这里有两类 - 信息可以为你提供参考 - - 1 'input fields after batch...'这部分显示的是train dataset经过Batch操作后，每个field对应的类型以及进行shape。这里 - 因为train dataset没有target所以没有显示。根据这里可以看出是否正确将需要的内容设置为了input或target。 - - 2 NameError，NameError发生在映射出错的情况。这里报错的原因是由于尝试进行forward计算时(可以通过Model.forward(self, x, b)判断 - 出当前是在调取forward)，却没有获取到forward()函数中需要的'x'；在报错信息中同时指出了缺'x'，而'a'没有被使用，那么可能 - 就是由于field的名称不对。这里将dataset中'a'这个field的名称改为'x'，或者model的参数从'x'修改为'a'都可以解决问题。 - - 下面的例子是由于loss计算的时候找不到需要的值 +---------------------------- + +.. code-block:: python + + import numpy as np + from torch import nn + import torch + from torch.optim import SGD + from fastNLP import Trainer + from fastNLP import DataSet + + class Model(nn.Module): + def __init__(self): + super().__init__() + self.fc = nn.Linear(1, 1) + def forward(self, x, b): + loss = torch.mean((self.fc(x)-b)**2) + return {'loss': loss} + model = Model() + + dataset = DataSet({'a': np.arange(10), 'b':np.arange(10)*2}) + dataset.set_input('a', 'b') + + trainer = Trainer(dataset, model, loss=None, optimizer=SGD(model.parameters(), lr=0.001)) + + trainer = Trainer(dataset, model, SGD(model.parameters())) + # 会报以下的错误 + # input fields after batch(if batch size is 2): + # a: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) + # b: (1)type:torch.Tensor (2)dtype:torch.int64, (3)shape:torch.Size([2]) + # There is no target field. + # .... + # NameError: + # Problems occurred when calling Model.forward(self, x, b) + # missing param: ['x'] + # unused field: ['a'] + # Suggestion: You need to provide ['x'] in DataSet and set it as input. + +这里就是由于在Trainer初始化的时候，fastNLP会尝试使用一个batch_size=2的batch去运行一遍forward()以及backward()。这里有两类 +信息可以为你提供参考 + +1 'input fields after batch...'这部分显示的是train dataset经过Batch操作后，每个field对应的类型以及进行shape。这里 +因为train dataset没有target所以没有显示。根据这里可以看出是否正确将需要的内容设置为了input或target。 + +2 NameError，NameError发生在映射出错的情况。这里报错的原因是由于尝试进行forward计算时(可以通过Model.forward(self, x, b)判断 +出当前是在调取forward)，却没有获取到forward()函数中需要的'x'；在报错信息中同时指出了缺'x'，而'a'没有被使用，那么可能 +就是由于field的名称不对。这里将dataset中'a'这个field的名称改为'x'，或者model的参数从'x'修改为'a'都可以解决问题。 + +下面的例子是由于loss计算的时候找不到需要的值 Example2.2 - :: - - import numpy as np - from torch import nn - from torch.optim import SGD - from fastNLP import Trainer - from fastNLP import DataSet - from fastNLP import L1Loss - import torch - - class Model(nn.Module): - def __init__(self): - super().__init__() - self.fc = nn.Linear(1, 1) - def forward(self, a): - return {'pred_b': self.fc(a.unsqueeze(1)).squeeze(1), 'No use':1} - - model = Model() - - dataset = DataSet({'a': np.arange(10, dtype=float), 'b':np.arange(10, dtype=float)*2}) - - dataset.set_input('a') - dataset.set_target('b') - - trainer = Trainer(dataset, model, loss=L1Loss(target='label'), optimizer=SGD(model.parameters(), lr=0.001)) - # 报错信息如下 - # input fields after batch(if batch size is 2): - # a: (1)type:torch.Tensor (2)dtype:torch.float32, (3)shape:torch.Size([2]) - # target fields after batch(if batch size is 2): - # b: (1)type:torch.Tensor (2)dtype:torch.float32, (3)shape:torch.Size([2]) - # .... - # NameError: - # Problems occurred when calling L1Loss.get_loss(self, pred, target) - # missing param: ['pred(assign to `pred` in `L1Loss`)', 'label(assign to `target` in `L1Loss`)'] - # unused field: ['b'] - # unused param: ['pred_b', 'No use'] - # target field: ['b'] - # param from Model.forward(self, a): ['pred_b', 'No use'] - # Suggestion: (1). Check key assignment for `target` when initialize L1Loss. Or provide `label` in DataSet or output of Model.forward(self, a). - # (2). Check key assignment for `pred` when initialize L1Loss. Or provide `pred` in DataSet or output of Model.forward(self, a). - - 报错信息也包含两部分: - - 1 第一部分与上面是一样的 - - 2 这里报错的原因是由于计算loss的时候找不到相应的值(通过L1Loss.get_loss(self, pred, target)判断出来的)； - 报错的原因是因为 `pred` 和 `label` (我们在初始化L1Loss时将target指定为了label)都没有找到。 - 这里'unused field'是DataSet中出现了，但却没有被设置为input或者target的field； - 'unused param'是forward()中返回且没有被使用到的内容；'target field'是被设置为了target的field; - 'param from Model.forward(self, a)'是forward()返回的所有key。"Suggestion"是关于当前错误处理的建议。 - - 但是在一些情况下，比如forward()返回值只有一个，target也只有一个，fastNLP不会进行匹配，而直接将forward()的结果作为pred, - 将DataSet中的target设置为target。上面的例子在返回值中加入了一个'No use'则只是为了使得Loss去匹配结果。 - - - 下面是带有dev dataset时如果出现错误会发生的报错， +---------------------------- + +.. code-block:: python + + import numpy as np + from torch import nn + from torch.optim import SGD + from fastNLP import Trainer + from fastNLP import DataSet + from fastNLP import L1Loss + import torch + + class Model(nn.Module): + def __init__(self): + super().__init__() + self.fc = nn.Linear(1, 1) + def forward(self, a): + return {'pred_b': self.fc(a.unsqueeze(1)).squeeze(1), 'No use':1} + + model = Model() + + dataset = DataSet({'a': np.arange(10, dtype=float), 'b':np.arange(10, dtype=float)*2}) + + dataset.set_input('a') + dataset.set_target('b') + + trainer = Trainer(dataset, model, loss=L1Loss(target='label'), optimizer=SGD(model.parameters(), lr=0.001)) + # 报错信息如下 + # input fields after batch(if batch size is 2): + # a: (1)type:torch.Tensor (2)dtype:torch.float32, (3)shape:torch.Size([2]) + # target fields after batch(if batch size is 2): + # b: (1)type:torch.Tensor (2)dtype:torch.float32, (3)shape:torch.Size([2]) + # .... + # NameError: + # Problems occurred when calling L1Loss.get_loss(self, pred, target) + # missing param: ['pred(assign to `pred` in `L1Loss`)', 'label(assign to `target` in `L1Loss`)'] + # unused field: ['b'] + # unused param: ['pred_b', 'No use'] + # target field: ['b'] + # param from Model.forward(self, a): ['pred_b', 'No use'] + # Suggestion: (1). Check key assignment for `target` when initialize L1Loss. Or provide `label` in DataSet or output of Model.forward(self, a). + # (2). Check key assignment for `pred` when initialize L1Loss. Or provide `pred` in DataSet or output of Model.forward(self, a). + +报错信息也包含两部分: + +1 第一部分与上面是一样的 + +2 这里报错的原因是由于计算loss的时候找不到相应的值(通过L1Loss.get_loss(self, pred, target)判断出来的)； +报错的原因是因为 `pred` 和 `label` (我们在初始化L1Loss时将target指定为了label)都没有找到。 +这里'unused field'是DataSet中出现了，但却没有被设置为input或者target的field； +'unused param'是forward()中返回且没有被使用到的内容；'target field'是被设置为了target的field; +'param from Model.forward(self, a)'是forward()返回的所有key。"Suggestion"是关于当前错误处理的建议。 + +但是在一些情况下，比如forward()返回值只有一个，target也只有一个，fastNLP不会进行匹配，而直接将forward()的结果作为pred, +将DataSet中的target设置为target。上面的例子在返回值中加入了一个'No use'则只是为了使得Loss去匹配结果。 + + +下面是带有dev dataset时如果出现错误会发生的报错， Example2.3 - :: +---------------------------- + +.. code-block:: python + + import numpy as np + from torch import nn + from torch.optim import SGD + from fastNLP import Trainer + from fastNLP import DataSet + from fastNLP import AccuracyMetric + import torch + + class Model(nn.Module): + def __init__(self): + super().__init__() + self.fc = nn.Linear(1, 1) + def forward(self, a, b): + loss = torch.mean((self.fc(a.float().unsqueeze(1))-b.float())**2) + return {'loss': loss} + def predict(self, a): # 使用predict()进行验证 + return {'output':self.fc(a.float().unsqueeze(1))} #这里return的值不包含'pred'这个key + model = Model() + + dataset = DataSet({'a': np.arange(10), 'b':np.arange(10)*2}) + dev_data = DataSet({'a': np.arange(10, 20), 'b':np.arange(10, 20)*2}) + + dataset.set_input('a', 'b') + dev_data.set_input('a') # 这里没有设置target + + trainer = Trainer(dataset, model, loss=None, optimizer=SGD(model.parameters(), lr=0.001), + dev_data=dev_data, metrics=AccuracyMetric()) + + # 报错信息 + # ... + # NameError: + # Problems occurred when calling AccuracyMetric.evaluate(self, pred, target, seq_len=None) + # missing param: ['pred(assign to `pred` in `AccuracyMetric`)', 'target(assign to `target` in `AccuracyMetric`)'] + # unused param: ['output'] + # target field: [] + # param from Model.predict(self, a): ['output'] + # Suggestion: (1). Check key assignment for `pred` when initialize AccuracyMetric. Or provide `pred` in DataSet or output of Model.predict(self, a). + # (2). Check key assignment for `target` when initialize AccuracyMetric. Or provide `target` in DataSet or output of Model.predict(self, a). + +报错信息和前面都是类似的，但是可以通过'AccuracyMetric.evaluate(self, pred, target, seq_len=None)'看出这里是evaluation +的时候发生了错误。这样避免了需要在完成一整个epoch的训练才能发现evaluation弄错的情况。这里的修改是通过在初始化metric的时候 +指明通过'output'获取`pred`, 即AccuracyMetric(pred='output')。 + +可以通过check_code_level调节检查的强度。默认为0，即进行检查。 + +---------------------------- +3. Trainer与callback +---------------------------- + +虽然Trainer本身已经集成了一些功能，但仍然不足以囊括训练过程中可能需要到的功能，比如负采样，learning rate decay, Early Stop等。 +为了解决这个问题fastNLP引入了callback的机制，:class:`~fastNLP.Callback` 是一种在Trainer训练过程中特定阶段会运行的函数集合， +所有的 :class:`~fastNLP.Callback` 都具有on_*(比如on_train_start, on_backward_begin)等函数。 +如果 Callback 实现了该函数，则Trainer运行至对应阶段，会进行调用，例如:: + + from fastNLP import Callback, EarlyStopCallback, Trainer, CrossEntropyLoss, AccuracyMetric + from fastNLP.models import CNNText + + start_time = time.time() - import numpy as np - from torch import nn - from torch.optim import SGD - from fastNLP import Trainer - from fastNLP import DataSet - from fastNLP import AccuracyMetric - import torch - - class Model(nn.Module): - def __init__(self): - super().__init__() - self.fc = nn.Linear(1, 1) - def forward(self, a, b): - loss = torch.mean((self.fc(a.float().unsqueeze(1))-b.float())**2) - return {'loss': loss} - def predict(self, a): # 使用predict()进行验证 - return {'output':self.fc(a.float().unsqueeze(1))} #这里return的值不包含'pred'这个key - model = Model() - - dataset = DataSet({'a': np.arange(10), 'b':np.arange(10)*2}) - dev_data = DataSet({'a': np.arange(10, 20), 'b':np.arange(10, 20)*2}) - - dataset.set_input('a', 'b') - dev_data.set_input('a') # 这里没有设置target - - trainer = Trainer(dataset, model, loss=None, optimizer=SGD(model.parameters(), lr=0.001), - dev_data=dev_data, metrics=AccuracyMetric()) - - # 报错信息 - # ... - # NameError: - # Problems occurred when calling AccuracyMetric.evaluate(self, pred, target, seq_len=None) - # missing param: ['pred(assign to `pred` in `AccuracyMetric`)', 'target(assign to `target` in `AccuracyMetric`)'] - # unused param: ['output'] - # target field: [] - # param from Model.predict(self, a): ['output'] - # Suggestion: (1). Check key assignment for `pred` when initialize AccuracyMetric. Or provide `pred` in DataSet or output of Model.predict(self, a). - # (2). Check key assignment for `target` when initialize AccuracyMetric. Or provide `target` in DataSet or output of Model.predict(self, a). - - 报错信息和前面都是类似的，但是可以通过'AccuracyMetric.evaluate(self, pred, target, seq_len=None)'看出这里是evaluation - 的时候发生了错误。这样避免了需要在完成一整个epoch的训练才能发现evaluation弄错的情况。这里的修改是通过在初始化metric的时候 - 指明通过'output'获取`pred`, 即AccuracyMetric(pred='output')。 - - 可以通过check_code_level调节检查的强度。默认为0，即进行检查。 - -3 Trainer与callback - 虽然Trainer本身已经集成了一些功能，但仍然不足以囊括训练过程中可能需要到的功能，比如负采样，learning rate decay, Early Stop等。 - 为了解决这个问题fastNLP引入了callback的机制，:class:`~fastNLP.Callback` 是一种在Trainer训练过程中特定阶段会运行的函数集合， - 所有的 :class:`~fastNLP.Callback` 都具有on_*(比如on_train_start, on_backward_begin)等函数。 - 如果 Callback 实现了该函数，则Trainer运行至对应阶段，会进行调用，例如:: + class MyCallback(Callback): + def on_epoch_end(self): + print('{:d}ms\n\n'.format(round((time.time()-start_time)*1000))) - from fastNLP import Callback, EarlyStopCallback, Trainer, CrossEntropyLoss, AccuracyMetric - from fastNLP.models import CNNText - - start_time = time.time() - - class MyCallback(Callback): - def on_epoch_end(self): - print('{:d}ms\n\n'.format(round((time.time()-start_time)*1000))) - - model = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) - trainer = Trainer(model=model, train_data=train_data, dev_data=dev_data, loss=CrossEntropyLoss(), - metrics=AccuracyMetric(), callbacks=[MyCallback(),EarlyStopCallback(10)]) - trainer.train() - - 这里，我们通过继承 :class:`~fastNLP.Callback` 类定义了自己的 callback 的，并和内置的 :class:`~fastNLP.EarlyStopCallback` - 一起传给了 :class:`~fastNLP.Trainer` ，增强了 :class:`~fastNLP.Trainer` 的功能 + model = CNNText((len(vocab),50), num_classes=5, padding=2, dropout=0.1) + trainer = Trainer(model=model, train_data=train_data, dev_data=dev_data, loss=CrossEntropyLoss(), + metrics=AccuracyMetric(), callbacks=[MyCallback(),EarlyStopCallback(10)]) + trainer.train() - fastNLP已经自带了很多callback函数供使用，可以参考 :doc:`fastNLP.core.callback` 。 +这里，我们通过继承 :class:`~fastNLP.Callback` 类定义了自己的 callback 的，并和内置的 :class:`~fastNLP.EarlyStopCallback` +一起传给了 :class:`~fastNLP.Trainer` ，增强了 :class:`~fastNLP.Trainer` 的功能 + +fastNLP已经自带了很多callback函数供使用，可以参考 :doc:`fastNLP.core.callback` 。 """ __all__ = [ @@ -330,6 +352,7 @@ from .utils import _move_dict_value_to_device from .utils import _get_func_signature from .utils import _get_model_device from .utils import _move_model_to_device +from ._parallel_utils import _model_contains_inner_module class Trainer(object): @@ -367,8 +390,8 @@ class Trainer(object): 要指定以哪个指标为准。另外有些指标是越小效果越好，比如语言模型的困惑度，这种情况下，在key前面增加一个'-'来表 明验证时，值越小越好(比如: "-ppl")。仅在传入dev_data时有效。 :param int validate_every: 多少个step在验证集上验证一次; 如果为-1，则每个epoch结束验证一次。仅在传入dev_data时有效。 - :param str,None save_path: 将模型保存路径。如果为None，则不保存模型。如果dev_data为None，则保存最后一次迭代的模型。 - 保存的时候不仅保存了参数，还保存了模型结构。即便使用DataParallel，这里也只保存模型。 + :param str,None save_path: 将模型保存路径，如果路径不存在，将自动创建文件夹。如果为None，则不保存模型。如果dev_data为None，则保存 + 最后一次迭代的模型。保存的时候不仅保存了参数，还保存了模型结构。即便使用DataParallel，这里也只保存模型。 :param bool use_tqdm: 是否使用tqdm来显示训练进度; 如果为False，则将loss打印在终端中。 :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 的计算位置进行管理。支持以下的输入: @@ -408,23 +431,23 @@ class Trainer(object): super(Trainer, self).__init__() if not isinstance(model, nn.Module): raise TypeError(f"The type of model must be torch.nn.Module, got {type(model)}.") - + # check metrics and dev_data if (not metrics) and dev_data is not None: raise ValueError("No metric for dev_data evaluation.") if metrics and (dev_data is None): raise ValueError("No dev_data for evaluations, pass dev_data or set metrics to None. ") - + # check update every assert update_every >= 1, "update_every must be no less than 1." self.update_every = int(update_every) - + # check save_path if not (save_path is None or isinstance(save_path, str)): raise ValueError("save_path can only be None or `str`.") # prepare evaluate metrics = _prepare_metrics(metrics) - + # parse metric_key # increase_better is True. It means the exp result gets better if the indicator increases. # It is true by default. @@ -436,28 +459,69 @@ class Trainer(object): self.metric_key = None # prepare loss losser = _prepare_losser(loss) - - # sampler check - if sampler is not None and not isinstance(sampler, Sampler): - raise ValueError("The type of sampler should be fastNLP.BaseSampler, got {}.".format(type(sampler))) - if sampler is None: - sampler = RandomSampler() + if isinstance(train_data, BatchIter): + if sampler is not None: + warnings.warn("sampler is ignored when train_data is a BatchIter.") + if num_workers>0: + warnings.warn("num_workers is ignored when train_data is BatchIter.") + if drop_last: + warnings.warn("drop_last is ignored when train_data is BatchIter.") + + if isinstance(model, nn.parallel.DistributedDataParallel): # 如果是分布式的 + # device为None + if device is not None: + warnings.warn("device is ignored when model is nn.parallel.DistributedDataParallel.") + device = None + # Sampler要是分布式的 + if sampler is None: + sampler = torch.utils.data.DistributedSampler(train_data) + elif not isinstance(sampler, torch.utils.data.DistributedSampler): + raise TypeError("When using nn.parallel.DistributedDataParallel, " + "sampler must be None or torch.utils.data.DistributedSampler.") + # 不能保存模型 + if save_path: + raise RuntimeError("Saving model in Distributed situation is not allowed right now.") + else: + # sampler check + if sampler is not None and not isinstance(sampler, (Sampler, torch.utils.data.Sampler)): + raise ValueError(f"The type of sampler should be fastNLP.BaseSampler or pytorch's Sampler, got {type(sampler)}") + if sampler is None: + sampler = RandomSampler() + elif hasattr(sampler, 'set_batch_size'): + sampler.set_batch_size(batch_size) if isinstance(train_data, DataSet): self.data_iterator = DataSetIter( dataset=train_data, batch_size=batch_size, num_workers=num_workers, sampler=sampler, drop_last=drop_last) elif isinstance(train_data, BatchIter): self.data_iterator = train_data + train_data = train_data.dataset else: raise TypeError("train_data type {} not support".format(type(train_data))) - if check_code_level > -1 and isinstance(self.data_iterator, DataSetIter): - _check_code(dataset=train_data, model=model, losser=losser, metrics=metrics, dev_data=dev_data, - metric_key=self.metric_key, check_level=check_code_level, - batch_size=min(batch_size, DEFAULT_CHECK_BATCH_SIZE)) - # _check_code 是 fastNLP 帮助你检查代码是否正确的方法 。如果你在错误栈中看到这行注释，请认真检查你的代码 self.model = _move_model_to_device(model, device=device) + if _model_contains_inner_module(self.model): + self._forward_func = self.model.module.forward + else: + self._forward_func = self.model.forward + if check_code_level > -1: + # _check_code 是 fastNLP 帮助你检查代码是否正确的方法 。如果你在错误栈中看到这行注释，请认真检查你的field名与模型的输入 + # 名是否匹配 + dev_dataset = dev_data + if isinstance(dev_data, BatchIter): + dev_dataset = None + warnings.warn("dev_data is of BatchIter type, ignore validation checking.") + check_batch_size = min(batch_size, DEFAULT_CHECK_BATCH_SIZE) + if isinstance(self.model, nn.DataParallel): + _num_devices = len(self.model.device_ids) + if batch_size//_num_devices>1: # 如果多卡是每个卡可以分多个数据的，则用每个卡给两个sample + check_batch_size = max(len(self.model.device_ids)*2, check_batch_size) + else: + check_batch_size = max(len(self.model.device_ids), check_batch_size) + _check_code(dataset=train_data, model=self.model, losser=losser, forward_func=self._forward_func, metrics=metrics, + dev_data=dev_dataset, metric_key=self.metric_key, check_level=check_code_level, + batch_size=check_batch_size) self.train_data = train_data self.dev_data = dev_data # If None, No validation. @@ -472,8 +536,7 @@ class Trainer(object): self.best_dev_epoch = None self.best_dev_step = None self.best_dev_perf = None - self.n_steps = (len(self.train_data) // self.batch_size + int( - len(self.train_data) % self.batch_size != 0)) * int(drop_last==0) * self.n_epochs + self.n_steps = len(self.data_iterator) * self.n_epochs if isinstance(optimizer, torch.optim.Optimizer): self.optimizer = optimizer @@ -483,7 +546,7 @@ class Trainer(object): self.optimizer = torch.optim.Adam(self.model.parameters(), lr=4e-3) else: raise TypeError("optimizer can only be torch.optim.Optimizer type, not {}.".format(type(optimizer))) - + self.use_tqdm = use_tqdm self.pbar = None self.print_every = abs(self.print_every) @@ -494,11 +557,12 @@ class Trainer(object): metrics=self.metrics, batch_size=self.batch_size, device=None, # 由上面的部分处理device - verbose=0) - + verbose=0, + use_tqdm=self.use_tqdm) + self.step = 0 self.start_time = None # start timestamp - + self.callback_manager = CallbackManager(env={"trainer": self}, callbacks=callbacks) @@ -534,7 +598,7 @@ class Trainer(object): self.start_time = str(datetime.now().strftime('%Y-%m-%d-%H-%M-%S')) start_time = time.time() print("training epochs started " + self.start_time, flush=True) - + try: self.callback_manager.on_train_begin() self._train() @@ -547,7 +611,7 @@ class Trainer(object): raise e elif on_exception == 'raise': raise e - + if self.dev_data is not None and self.best_dev_perf is not None: print( "\nIn Epoch:{}/Step:{}, got best dev performance:".format(self.best_dev_epoch, self.best_dev_step) + @@ -565,21 +629,17 @@ class Trainer(object): finally: pass results['seconds'] = round(time.time() - start_time, 2) - + return results - + def _train(self): if not self.use_tqdm: - from fastNLP.core.utils import _pseudo_tqdm as inner_tqdm + from .utils import _pseudo_tqdm as inner_tqdm else: inner_tqdm = tqdm self.step = 0 self.epoch = 0 start = time.time() - if isinstance(self.model, nn.DataParallel): - self._forward_func = self.model.module.forward - else: - self._forward_func = self.model.forward with inner_tqdm(total=self.n_steps, postfix='loss:{0:<6.5f}', leave=False, dynamic_ncols=True) as pbar: self.pbar = pbar avg_loss = 0 @@ -597,21 +657,21 @@ class Trainer(object): # negative sampling; replace unknown; re-weight batch_y self.callback_manager.on_batch_begin(batch_x, batch_y, indices) prediction = self._data_forward(self.model, batch_x) - + # edit prediction self.callback_manager.on_loss_begin(batch_y, prediction) loss = self._compute_loss(prediction, batch_y).mean() avg_loss += loss.item() loss = loss / self.update_every - + # Is loss NaN or inf? requires_grad = False self.callback_manager.on_backward_begin(loss) self._grad_backward(loss) self.callback_manager.on_backward_end() - + self._update() self.callback_manager.on_step_end() - + if self.step % self.print_every == 0: avg_loss = float(avg_loss) / self.print_every if self.use_tqdm: @@ -625,7 +685,7 @@ class Trainer(object): pbar.set_postfix_str(print_output) avg_loss = 0 self.callback_manager.on_batch_end() - + if ((self.validate_every > 0 and self.step % self.validate_every == 0) or (self.validate_every < 0 and self.step % len(data_iterator) == 0)) \ and self.dev_data is not None: @@ -634,20 +694,20 @@ class Trainer(object): self.n_steps) + \ self.tester._format_eval_results(eval_res) pbar.write(eval_str + '\n') - + # ================= mini-batch end ==================== # - + # lr decay; early stopping self.callback_manager.on_epoch_end() # =============== epochs end =================== # pbar.close() self.pbar = None # ============ tqdm end ============== # - + def _do_validation(self, epoch, step): self.callback_manager.on_valid_begin() res = self.tester.test() - + is_better_eval = False if self._better_eval_result(res): if self.save_path is not None: @@ -662,7 +722,7 @@ class Trainer(object): # get validation results; adjust optimizer self.callback_manager.on_valid_end(res, self.metric_key, self.optimizer, is_better_eval) return res - + def _mode(self, model, is_test=False): """Train mode or Test mode. This is for PyTorch currently. @@ -674,14 +734,14 @@ class Trainer(object): model.eval() else: model.train() - + def _update(self): """Perform weight update on a model. """ if self.step % self.update_every == 0: self.optimizer.step() - + def _data_forward(self, network, x): x = _build_args(self._forward_func, **x) y = network(**x) @@ -689,7 +749,7 @@ class Trainer(object): raise TypeError( f"The return value of {_get_func_signature(self._forward_func)} should be dict, got {type(y)}.") return y - + def _grad_backward(self, loss): """Compute gradient with link rules. @@ -700,7 +760,7 @@ class Trainer(object): if (self.step-1) % self.update_every == 0: self.model.zero_grad() loss.backward() - + def _compute_loss(self, predict, truth): """Compute loss given prediction and ground truth. @@ -709,7 +769,7 @@ class Trainer(object): :return: a scalar """ return self.losser(predict, truth) - + def _save_model(self, model, model_name, only_param=False): """ 存储不含有显卡信息的state_dict或model :param model: @@ -721,7 +781,7 @@ class Trainer(object): model_path = os.path.join(self.save_path, model_name) if not os.path.exists(self.save_path): os.makedirs(self.save_path, exist_ok=True) - if isinstance(model, nn.DataParallel): + if _model_contains_inner_module(model): model = model.module if only_param: state_dict = model.state_dict() @@ -732,7 +792,7 @@ class Trainer(object): model.cpu() torch.save(model, model_path) model.to(self._model_device) - + def _load_model(self, model, model_name, only_param=False): # 返回bool值指示是否成功reload模型 if self.save_path is not None: @@ -741,7 +801,7 @@ class Trainer(object): states = torch.load(model_path) else: states = torch.load(model_path).state_dict() - if isinstance(model, nn.DataParallel): + if _model_contains_inner_module(model): model.module.load_state_dict(states) else: model.load_state_dict(states) @@ -750,7 +810,7 @@ class Trainer(object): else: return False return True - + def _better_eval_result(self, metrics): """Check if the current epoch yields better validation results. @@ -776,6 +836,9 @@ class Trainer(object): is_better = False return is_better + @property + def is_master(self): + return True DEFAULT_CHECK_BATCH_SIZE = 2 DEFAULT_CHECK_NUM_BATCH = 2 @@ -797,14 +860,15 @@ def _get_value_info(_dict): strs.append(_str) return strs + from numbers import Number from .batch import _to_tensor -def _check_code(dataset, model, losser, metrics, batch_size=DEFAULT_CHECK_BATCH_SIZE, - dev_data=None, metric_key=None, - check_level=0): + + +def _check_code(dataset, model, losser, metrics, forward_func, batch_size=DEFAULT_CHECK_BATCH_SIZE, + dev_data=None, metric_key=None, check_level=0): # check get_loss 方法 - model_devcie = _get_model_device(model=model) - + model_device = _get_model_device(model=model) def _iter(): start_idx = 0 while start_idx>> seq_len = torch.arange(2, 16) >>> mask = seq_len_to_mask(seq_len) @@ -691,7 +634,7 @@ def seq_len_to_mask(seq_len, max_len=None): :param np.ndarray,torch.LongTensor seq_len: shape将是(B,) :param int max_len: 将长度pad到这个长度。默认(None)使用的是seq_len中最长的长度。但在nn.DataParallel的场景下可能不同卡的seq_len会有 区别，所以需要传入一个max_len使得mask的长度是pad到该长度。 - :return: np.ndarray or torch.Tensor, shape将是(B, max_length)。 元素类似为bool或torch.uint8 + :return: np.ndarray, torch.Tensor 。shape将是(B, max_length)， 元素类似为bool或torch.uint8 """ if isinstance(seq_len, np.ndarray): assert len(np.shape(seq_len)) == 1, f"seq_len can only have one dimension, got {len(np.shape(seq_len))}." @@ -737,7 +680,8 @@ class _pseudo_tqdm: def __exit__(self, exc_type, exc_val, exc_tb): del self -def iob2(tags:List[str])->List[str]: + +def iob2(tags: List[str]) -> List[str]: """ 检查数据是否是合法的IOB数据，如果是IOB1会被自动转换为IOB2。两者的差异见 https://datascience.stackexchange.com/questions/37824/difference-between-iob-and-iob2-format @@ -760,7 +704,8 @@ def iob2(tags:List[str])->List[str]: tags[i] = "B" + tag[1:] return tags -def iob2bioes(tags:List[str])->List[str]: + +def iob2bioes(tags: List[str]) -> List[str]: """ 将iob的tag转换为bioes编码 :param tags: List[str]. 编码需要是大写的。 @@ -773,15 +718,15 @@ def iob2bioes(tags:List[str])->List[str]: else: split = tag.split('-')[0] if split == 'B': - if i+1!=len(tags) and tags[i+1].split('-')[0] == 'I': + if i + 1 != len(tags) and tags[i + 1].split('-')[0] == 'I': new_tags.append(tag) else: new_tags.append(tag.replace('B-', 'S-')) elif split == 'I': - if i + 1>> import torch + >>> from fastNLP import Vocabulary + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> embed = BertEmbedding(vocab, model_dir_or_name='en-base-uncased', requires_grad=False, layers='4,-2,-1') + >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) + >>> outputs = embed(words) + >>> outputs.size() + >>> # torch.Size([1, 5, 2304]) + + :param ~fastNLP.Vocabulary vocab: 词表 + :param str model_dir_or_name: 模型所在目录或者模型的名称。当传入模型所在目录时，目录中应该包含一个词表文件(以.txt作为后缀名), + 权重文件(以.bin作为文件后缀名), 配置文件(以.json作为后缀名)。 + :param str layers: 输出embedding表示来自于哪些层，不同层的结果按照layers中的顺序在最后一维concat起来。以','隔开层数，层的序号是 + 从0开始，可以以负数去索引倒数几层。 + :param str pool_method: 因为在bert中，每个word会被表示为多个word pieces, 当获取一个word的表示的时候，怎样从它的word pieces + 中计算得到它对应的表示。支持 ``last`` , ``first`` , ``avg`` , ``max``。 + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 + :param bool include_cls_sep: bool，在bert计算句子的表示的时候，需要在前面加上[CLS]和[SEP], 是否在结果中保留这两个内容。 这样 + 会使得word embedding的结果比输入的结果长两个token。如果该值为True，则在使用 :class::StackEmbedding 可能会与其它类型的 + embedding长度不匹配。 + :param bool pooled_cls: 返回的[CLS]是否使用预训练中的BertPool映射一下，仅在include_cls_sep时有效。如果下游任务只取[CLS]做预测， + 一般该值为True。 + :param bool requires_grad: 是否需要gradient以更新Bert的权重。 + :param bool auto_truncate: 当句子words拆分为word pieces长度超过bert最大允许长度(一般为512), 自动截掉拆分后的超过510个 + word pieces后的内容，并将第512个word piece置为[SEP]。超过长度的部分的encode结果直接全部置零。一般仅有只使用[CLS] + 来进行分类的任务将auto_truncate置为True。 + """ + def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en-base-uncased', layers: str='-1', + pool_method: str='first', word_dropout=0, dropout=0, include_cls_sep: bool=False, + pooled_cls=True, requires_grad: bool=False, auto_truncate:bool=False): + super(BertEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + + # 根据model_dir_or_name检查是否存在并下载 + if model_dir_or_name.lower() in PRETRAINED_BERT_MODEL_DIR: + PRETRAIN_URL = _get_base_url('bert') + model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] + model_url = PRETRAIN_URL + model_name + model_dir = cached_path(model_url) + # 检查是否存在 + elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): + model_dir = os.path.expanduser(os.path.abspath(model_dir_or_name)) + else: + raise ValueError(f"Cannot recognize {model_dir_or_name}.") + + self.model = _WordBertModel(model_dir=model_dir, vocab=vocab, layers=layers, + pool_method=pool_method, include_cls_sep=include_cls_sep, + pooled_cls=pooled_cls, auto_truncate=auto_truncate) + + self.requires_grad = requires_grad + self._embed_size = len(self.model.layers)*self.model.encoder.hidden_size + + def _delete_model_weights(self): + del self.model + + def forward(self, words): + """ + 计算words的bert embedding表示。计算之前会在每句话的开始增加[CLS]在结束增加[SEP], 并根据include_cls_sep判断要不要 + 删除这两个token的表示。 + + :param torch.LongTensor words: [batch_size, max_len] + :return: torch.FloatTensor. batch_size x max_len x (768*len(self.layers)) + """ + words = self.drop_word(words) + outputs = self._get_sent_reprs(words) + if outputs is not None: + return self.dropout(words) + outputs = self.model(words) + outputs = torch.cat([*outputs], dim=-1) + + return self.dropout(outputs) + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + + :return: + """ + requires_grads = set([param.requires_grad for name, param in self.named_parameters() + if 'word_pieces_lengths' not in name]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + if 'word_pieces_lengths' in name: # 这个不能加入到requires_grad中 + continue + param.requires_grad = value + + +class BertWordPieceEncoder(nn.Module): + """ + 读取bert模型，读取之后调用index_dataset方法在dataset中生成word_pieces这一列。 + + :param str model_dir_or_name: 模型所在目录或者模型的名称。默认值为 ``en-base-uncased`` + :param str layers: 最终结果中的表示。以','隔开层数，可以以负数去索引倒数几层 + :param bool pooled_cls: 返回的句子开头的[CLS]是否使用预训练中的BertPool映射一下，仅在include_cls_sep时有效。如果下游任务只取 + [CLS]做预测，一般该值为True。 + :param bool requires_grad: 是否需要gradient。 + """ + def __init__(self, model_dir_or_name: str='en-base-uncased', layers: str='-1', + pooled_cls: bool = False, requires_grad: bool=False): + super().__init__() + + if model_dir_or_name in PRETRAINED_BERT_MODEL_DIR: + PRETRAIN_URL = _get_base_url('bert') + model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] + model_url = PRETRAIN_URL + model_name + model_dir = cached_path(model_url) + # 检查是否存在 + elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): + model_dir = model_dir_or_name + else: + raise ValueError(f"Cannot recognize {model_dir_or_name}.") + + self.model = _WordPieceBertModel(model_dir=model_dir, layers=layers, pooled_cls=pooled_cls) + self._embed_size = len(self.model.layers) * self.model.encoder.hidden_size + self.requires_grad = requires_grad + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + :return: + """ + requires_grads = set([param.requires_grad for name, param in self.named_parameters()]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + param.requires_grad = value + + @property + def embed_size(self): + return self._embed_size + + @property + def embedding_dim(self): + return self._embed_size + + @property + def num_embedding(self): + return self.model.encoder.config.vocab_size + + def index_datasets(self, *datasets, field_name, add_cls_sep=True): + """ + 使用bert的tokenizer新生成word_pieces列加入到datasets中，并将他们设置为input,且将word_pieces这一列的pad value设置为了 + bert的pad value。 + + :param DataSet datasets: DataSet对象 + :param str field_name: 基于哪一列的内容生成word_pieces列。这一列中每个数据应该是List[str]的形式。 + :param bool add_cls_sep: 如果首尾不是[CLS]与[SEP]会在首尾额外加入[CLS]与[SEP]。 + :return: + """ + self.model.index_dataset(*datasets, field_name=field_name, add_cls_sep=add_cls_sep) + + def forward(self, word_pieces, token_type_ids=None): + """ + 计算words的bert embedding表示。传入的words中应该自行包含[CLS]与[SEP]的tag。 + + :param words: batch_size x max_len + :param token_type_ids: batch_size x max_len, 用于区分前一句和后一句话 + :return: torch.FloatTensor. batch_size x max_len x (768*len(self.layers)) + """ + outputs = self.model(word_pieces, token_type_ids) + outputs = torch.cat([*outputs], dim=-1) + + return outputs + + +class _WordBertModel(nn.Module): + def __init__(self, model_dir:str, vocab:Vocabulary, layers:str='-1', pool_method:str='first', + include_cls_sep:bool=False, pooled_cls:bool=False, auto_truncate:bool=False): + super().__init__() + + self.tokenzier = BertTokenizer.from_pretrained(model_dir) + self.encoder = BertModel.from_pretrained(model_dir) + self._max_position_embeddings = self.encoder.config.max_position_embeddings + # 检查encoder_layer_number是否合理 + encoder_layer_number = len(self.encoder.encoder.layer) + self.layers = list(map(int, layers.split(','))) + for layer in self.layers: + if layer<0: + assert -layer<=encoder_layer_number, f"The layer index:{layer} is out of scope for " \ + f"a bert model with {encoder_layer_number} layers." + else: + assert layerself._max_position_embeddings: + if self.auto_truncate: + word_pieces_lengths = word_pieces_lengths.masked_fill(word_pieces_lengths+2>self._max_position_embeddings, + self._max_position_embeddings-2) + max_word_piece_length = self._max_position_embeddings-2 + else: + raise RuntimeError("After split words into word pieces, the lengths of word pieces are longer than the " + f"maximum allowed sequence length:{self._max_position_embeddings} of bert.") + + + # +2是由于需要加入[CLS]与[SEP] + word_pieces = words.new_full((batch_size, max_word_piece_length+2), fill_value=self._wordpiece_pad_index) + word_pieces[:, 0].fill_(self._cls_index) + batch_indexes = torch.arange(batch_size).to(words) + word_pieces[batch_indexes, word_pieces_lengths+1] = self._sep_index + attn_masks = torch.zeros_like(word_pieces) + # 1. 获取words的word_pieces的id，以及对应的span范围 + word_indexes = words.tolist() + for i in range(batch_size): + word_pieces_i = list(chain(*self.word_to_wordpieces[word_indexes[i]])) + if self.auto_truncate and len(word_pieces_i)>self._max_position_embeddings-2: + word_pieces_i = word_pieces_i[:self._max_position_embeddings-2] + word_pieces[i, 1:len(word_pieces_i)+1] = torch.LongTensor(word_pieces_i) + attn_masks[i, :len(word_pieces_i)+2].fill_(1) + # TODO 截掉长度超过的部分。 + # 2. 获取hidden的结果，根据word_pieces进行对应的pool计算 + # all_outputs: [batch_size x max_len x hidden_size, batch_size x max_len x hidden_size, ...] + bert_outputs, pooled_cls = self.encoder(word_pieces, token_type_ids=None, attention_mask=attn_masks, + output_all_encoded_layers=True) + # output_layers = [self.layers] # len(self.layers) x batch_size x max_word_piece_length x hidden_size + + if self.include_cls_sep: + outputs = bert_outputs[-1].new_zeros(len(self.layers), batch_size, max_word_len + 2, + bert_outputs[-1].size(-1)) + s_shift = 1 + else: + outputs = bert_outputs[-1].new_zeros(len(self.layers), batch_size, max_word_len, + bert_outputs[-1].size(-1)) + s_shift = 0 + batch_word_pieces_cum_length = batch_word_pieces_length.new_zeros(batch_size, max_word_len + 1) + batch_word_pieces_cum_length[:, 1:] = batch_word_pieces_length.cumsum(dim=-1) # batch_size x max_len + for l_index, l in enumerate(self.layers): + output_layer = bert_outputs[l] + if real_max_word_piece_length > max_word_piece_length: # 如果实际上是截取出来的 + paddings = output_layer.new_zeros(batch_size, + real_max_word_piece_length-max_word_piece_length, + output_layer.size(2)) + output_layer = torch.cat((output_layer, paddings), dim=1).contiguous() + # 从word_piece collapse到word的表示 + truncate_output_layer = output_layer[:, 1:-1] # 删除[CLS]与[SEP] batch_size x len x hidden_size + outputs_seq_len = seq_len + s_shift + if self.pool_method == 'first': + for i in range(batch_size): + i_word_pieces_cum_length = batch_word_pieces_cum_length[i, :seq_len[i]] # 每个word的start位置 + outputs[l_index, i, s_shift:outputs_seq_len[i]] = truncate_output_layer[i, i_word_pieces_cum_length] # num_layer x batch_size x len x hidden_size + elif self.pool_method == 'last': + for i in range(batch_size): + i_word_pieces_cum_length = batch_word_pieces_cum_length[i, 1:seq_len[i]+1] - 1 # 每个word的end + outputs[l_index, i, s_shift:outputs_seq_len[i]] = truncate_output_layer[i, i_word_pieces_cum_length] + elif self.pool_method == 'max': + for i in range(batch_size): + for j in range(seq_len[i]): + start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j+1] + outputs[l_index, i, j+s_shift], _ = torch.max(truncate_output_layer[i, start:end], dim=-2) + else: + for i in range(batch_size): + for j in range(seq_len[i]): + start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j+1] + outputs[l_index, i, j+s_shift] = torch.mean(truncate_output_layer[i, start:end], dim=-2) + if self.include_cls_sep: + if l in (len(bert_outputs)-1, -1) and self.pooled_cls: + outputs[l_index, :, 0] = pooled_cls + else: + outputs[l_index, :, 0] = output_layer[:, 0] + outputs[l_index, batch_indexes, seq_len+s_shift] = output_layer[batch_indexes, seq_len+s_shift] + # 3. 最终的embedding结果 + return outputs + diff --git a/fastNLP/embeddings/char_embedding.py b/fastNLP/embeddings/char_embedding.py new file mode 100644 index 00000000..b670313e --- /dev/null +++ b/fastNLP/embeddings/char_embedding.py @@ -0,0 +1,295 @@ +""" +该文件中主要包含的是character的Embedding，包括基于CNN与LSTM的character Embedding。与其它Embedding一样，这里的Embedding输入也是 +词的index而不需要使用词语中的char的index来获取表达。 +""" + + +import torch +import torch.nn as nn +import torch.nn.functional as F +from typing import List + +from ..modules.encoder.lstm import LSTM +from ..core.vocabulary import Vocabulary +from .embedding import TokenEmbedding +from .utils import _construct_char_vocab_from_vocab + + +class CNNCharEmbedding(TokenEmbedding): + """ + 别名：:class:`fastNLP.embeddings.CNNCharEmbedding` :class:`fastNLP.embeddings.char_embedding.CNNCharEmbedding` + + 使用CNN生成character embedding。CNN的结构为, embed(x) -> Dropout(x) -> CNN(x) -> activation(x) -> pool -> fc -> Dropout. + 不同的kernel大小的fitler结果是concat起来然后通过一层fully connected layer, 然后输出word的表示。 + + Example:: + + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> embed = CNNCharEmbedding(vocab, embed_size=50) + >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) + >>> outputs = embed(words) + >>> outputs.size() + >>> # torch.Size([1, 5，50]) + + :param vocab: 词表 + :param embed_size: 该word embedding的大小，默认值为50. + :param char_emb_size: character的embed的大小。character是从vocab中生成的。默认值为50. + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param float dropout: 以多大的概率drop分布式表示与char embedding的输出。 + :param filter_nums: filter的数量. 长度需要和kernels一致。默认值为[40, 30, 20]. + :param kernel_sizes: kernel的大小. 默认值为[5, 3, 1]. + :param pool_method: character的表示在合成一个表示时所使用的pool方法，支持'avg', 'max'. + :param activation: CNN之后使用的激活方法，支持'relu', 'sigmoid', 'tanh' 或者自定义函数. + :param min_char_freq: character的最少出现次数。默认值为2. + """ + def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, + dropout:float=0.5, filter_nums: List[int]=(40, 30, 20), kernel_sizes: List[int]=(5, 3, 1), + pool_method: str='max', activation='relu', min_char_freq: int=2): + super(CNNCharEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + + for kernel in kernel_sizes: + assert kernel % 2 == 1, "Only odd kernel is allowed." + + assert pool_method in ('max', 'avg') + self.dropout = nn.Dropout(dropout) + self.pool_method = pool_method + # activation function + if isinstance(activation, str): + if activation.lower() == 'relu': + self.activation = F.relu + elif activation.lower() == 'sigmoid': + self.activation = F.sigmoid + elif activation.lower() == 'tanh': + self.activation = F.tanh + elif activation is None: + self.activation = lambda x: x + elif callable(activation): + self.activation = activation + else: + raise Exception( + "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") + + print("Start constructing character vocabulary.") + # 建立char的词表 + self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) + self.char_pad_index = self.char_vocab.padding_idx + print(f"In total, there are {len(self.char_vocab)} distinct characters.") + # 对vocab进行index + max_word_len = max(map(lambda x: len(x[0]), vocab)) + self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), max_word_len), + fill_value=self.char_pad_index, dtype=torch.long), + requires_grad=False) + self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) + for word, index in vocab: + # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了。修改为不区分pad, 这样所有的也是同一个embed + self.words_to_chars_embedding[index, :len(word)] = \ + torch.LongTensor([self.char_vocab.to_index(c) for c in word]) + self.word_lengths[index] = len(word) + self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) + + self.convs = nn.ModuleList([nn.Conv1d( + char_emb_size, filter_nums[i], kernel_size=kernel_sizes[i], bias=True, padding=kernel_sizes[i] // 2) + for i in range(len(kernel_sizes))]) + self._embed_size = embed_size + self.fc = nn.Linear(sum(filter_nums), embed_size) + self.reset_parameters() + + def forward(self, words): + """ + 输入words的index后，生成对应的words的表示。 + + :param words: [batch_size, max_len] + :return: [batch_size, max_len, embed_size] + """ + words = self.drop_word(words) + batch_size, max_len = words.size() + chars = self.words_to_chars_embedding[words] # batch_size x max_len x max_word_len + word_lengths = self.word_lengths[words] # batch_size x max_len + max_word_len = word_lengths.max() + chars = chars[:, :, :max_word_len] + # 为1的地方为mask + chars_masks = chars.eq(self.char_pad_index) # batch_size x max_len x max_word_len 如果为0, 说明是padding的位置了 + chars = self.char_embedding(chars) # batch_size x max_len x max_word_len x embed_size + chars = self.dropout(chars) + reshaped_chars = chars.reshape(batch_size*max_len, max_word_len, -1) + reshaped_chars = reshaped_chars.transpose(1, 2) # B' x E x M + conv_chars = [conv(reshaped_chars).transpose(1, 2).reshape(batch_size, max_len, max_word_len, -1) + for conv in self.convs] + conv_chars = torch.cat(conv_chars, dim=-1).contiguous() # B x max_len x max_word_len x sum(filters) + conv_chars = self.activation(conv_chars) + if self.pool_method == 'max': + conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) + chars, _ = torch.max(conv_chars, dim=-2) # batch_size x max_len x sum(filters) + else: + conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), 0) + chars = torch.sum(conv_chars, dim=-2)/chars_masks.eq(0).sum(dim=-1, keepdim=True).float() + chars = self.fc(chars) + return self.dropout(chars) + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + :return: + """ + params = [] + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' not in name and 'word_lengths' not in name: + params.append(param.requires_grad) + requires_grads = set(params) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能加入到requires_grad中 + continue + param.requires_grad = value + + def reset_parameters(self): + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能reset + continue + if param.data.dim()>1: + nn.init.xavier_uniform_(param, 1) + else: + nn.init.uniform_(param, -1, 1) + + +class LSTMCharEmbedding(TokenEmbedding): + """ + 别名：:class:`fastNLP.embeddings.LSTMCharEmbedding` :class:`fastNLP.embeddings.char_embedding.LSTMCharEmbedding` + + 使用LSTM的方式对character进行encode. embed(x) -> Dropout(x) -> LSTM(x) -> activation(x) -> pool -> Dropout + + Example:: + + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> embed = LSTMCharEmbedding(vocab, embed_size=50) + >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) + >>> outputs = embed(words) + >>> outputs.size() + >>> # torch.Size([1, 5，50]) + + :param vocab: 词表 + :param embed_size: embedding的大小。默认值为50. + :param char_emb_size: character的embedding的大小。默认值为50. + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param dropout: 以多大概率drop character embedding的输出以及最终的word的输出。 + :param hidden_size: LSTM的中间hidden的大小，如果为bidirectional的，hidden会除二，默认为50. + :param pool_method: 支持'max', 'avg'。 + :param activation: 激活函数，支持'relu', 'sigmoid', 'tanh', 或者自定义函数. + :param min_char_freq: character的最小出现次数。默认值为2. + :param bidirectional: 是否使用双向的LSTM进行encode。默认值为True。 + """ + def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, + dropout:float=0.5, hidden_size=50,pool_method: str='max', activation='relu', min_char_freq: int=2, + bidirectional=True): + super(LSTMCharEmbedding, self).__init__(vocab) + + assert hidden_size % 2 == 0, "Only even kernel is allowed." + + assert pool_method in ('max', 'avg') + self.pool_method = pool_method + self.dropout = nn.Dropout(dropout) + # activation function + if isinstance(activation, str): + if activation.lower() == 'relu': + self.activation = F.relu + elif activation.lower() == 'sigmoid': + self.activation = F.sigmoid + elif activation.lower() == 'tanh': + self.activation = F.tanh + elif activation is None: + self.activation = lambda x: x + elif callable(activation): + self.activation = activation + else: + raise Exception( + "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") + + print("Start constructing character vocabulary.") + # 建立char的词表 + self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) + self.char_pad_index = self.char_vocab.padding_idx + print(f"In total, there are {len(self.char_vocab)} distinct characters.") + # 对vocab进行index + self.max_word_len = max(map(lambda x: len(x[0]), vocab)) + self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), self.max_word_len), + fill_value=self.char_pad_index, dtype=torch.long), + requires_grad=False) + self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) + for word, index in vocab: + # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了. 修改为不区分pad与否 + self.words_to_chars_embedding[index, :len(word)] = \ + torch.LongTensor([self.char_vocab.to_index(c) for c in word]) + self.word_lengths[index] = len(word) + self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) + + self.fc = nn.Linear(hidden_size, embed_size) + hidden_size = hidden_size // 2 if bidirectional else hidden_size + + self.lstm = LSTM(char_emb_size, hidden_size, bidirectional=bidirectional, batch_first=True) + self._embed_size = embed_size + self.bidirectional = bidirectional + + def forward(self, words): + """ + 输入words的index后，生成对应的words的表示。 + + :param words: [batch_size, max_len] + :return: [batch_size, max_len, embed_size] + """ + words = self.drop_word(words) + batch_size, max_len = words.size() + chars = self.words_to_chars_embedding[words] # batch_size x max_len x max_word_len + word_lengths = self.word_lengths[words] # batch_size x max_len + max_word_len = word_lengths.max() + chars = chars[:, :, :max_word_len] + # 为mask的地方为1 + chars_masks = chars.eq(self.char_pad_index) # batch_size x max_len x max_word_len 如果为0, 说明是padding的位置了 + chars = self.char_embedding(chars) # batch_size x max_len x max_word_len x embed_size + chars = self.dropout(chars) + reshaped_chars = chars.reshape(batch_size * max_len, max_word_len, -1) + char_seq_len = chars_masks.eq(0).sum(dim=-1).reshape(batch_size * max_len) + lstm_chars = self.lstm(reshaped_chars, char_seq_len)[0].reshape(batch_size, max_len, max_word_len, -1) + # B x M x M x H + + lstm_chars = self.activation(lstm_chars) + if self.pool_method == 'max': + lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) + chars, _ = torch.max(lstm_chars, dim=-2) # batch_size x max_len x H + else: + lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), 0) + chars = torch.sum(lstm_chars, dim=-2) / chars_masks.eq(0).sum(dim=-1, keepdim=True).float() + + chars = self.fc(chars) + + return self.dropout(chars) + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + + :return: + """ + params = [] + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' not in name and 'word_lengths' not in name: + params.append(param) + requires_grads = set(params) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能加入到requires_grad中 + continue + param.requires_grad = value diff --git a/fastNLP/embeddings/contextual_embedding.py b/fastNLP/embeddings/contextual_embedding.py new file mode 100644 index 00000000..1831af4e --- /dev/null +++ b/fastNLP/embeddings/contextual_embedding.py @@ -0,0 +1,100 @@ + +from abc import abstractmethod +import torch + +from ..core.vocabulary import Vocabulary +from ..core.dataset import DataSet +from ..core.batch import DataSetIter +from ..core.sampler import SequentialSampler +from ..core.utils import _move_model_to_device, _get_model_device +from .embedding import TokenEmbedding + + +class ContextualEmbedding(TokenEmbedding): + def __init__(self, vocab: Vocabulary, word_dropout:float=0.0, dropout:float=0.0): + super(ContextualEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + + def add_sentence_cache(self, *datasets, batch_size=32, device='cpu', delete_weights: bool=True): + """ + 由于动态embedding生成比较耗时，所以可以把每句话embedding缓存下来，这样就不需要每次都运行生成过程。 + + :param datasets: DataSet对象 + :param batch_size: int, 生成cache的sentence表示时使用的batch的大小 + :param device: 参考 :class::fastNLP.Trainer 的device + :param delete_weights: 似乎在生成了cache之后删除权重，在不需要finetune动态模型的情况下，删除权重会大量减少内存占用。 + :return: + """ + for index, dataset in enumerate(datasets): + try: + assert isinstance(dataset, DataSet), "Only fastNLP.DataSet object is allowed." + assert 'words' in dataset.get_input_name(), "`words` field has to be set as input." + except Exception as e: + print(f"Exception happens at {index} dataset.") + raise e + + sent_embeds = {} + _move_model_to_device(self, device=device) + device = _get_model_device(self) + pad_index = self._word_vocab.padding_idx + print("Start to calculate sentence representations.") + with torch.no_grad(): + for index, dataset in enumerate(datasets): + try: + batch = DataSetIter(dataset, batch_size=batch_size, sampler=SequentialSampler()) + for batch_x, batch_y in batch: + words = batch_x['words'].to(device) + words_list = words.tolist() + seq_len = words.ne(pad_index).sum(dim=-1) + max_len = words.size(1) + # 因为有些情况可能包含CLS, SEP, 从后面往前计算比较安全。 + seq_len_from_behind = (max_len - seq_len).tolist() + word_embeds = self(words).detach().cpu().numpy() + for b in range(words.size(0)): + length = seq_len_from_behind[b] + if length==0: + sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b] + else: + sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b, :-length] + except Exception as e: + print(f"Exception happens at {index} dataset.") + raise e + print("Finish calculating sentence representations.") + self.sent_embeds = sent_embeds + if delete_weights: + self._delete_model_weights() + + def _get_sent_reprs(self, words): + """ + 获取sentence的表示，如果有缓存，则返回缓存的值; 没有缓存则返回None + + :param words: torch.LongTensor + :return: + """ + if hasattr(self, 'sent_embeds'): + words_list = words.tolist() + seq_len = words.ne(self._word_pad_index).sum(dim=-1) + _embeds = [] + for b in range(len(words)): + words_i = tuple(words_list[b][:seq_len[b]]) + embed = self.sent_embeds[words_i] + _embeds.append(embed) + max_sent_len = max(map(len, _embeds)) + embeds = words.new_zeros(len(_embeds), max_sent_len, self.embed_size, dtype=torch.float, + device=words.device) + for i, embed in enumerate(_embeds): + embeds[i, :len(embed)] = torch.FloatTensor(embed).to(words.device) + return embeds + return None + + @abstractmethod + def _delete_model_weights(self): + """删除计算表示的模型以节省资源""" + raise NotImplementedError + + def remove_sentence_cache(self): + """ + 删除缓存的句子表示. 删除之后如果模型权重没有被删除，将开始使用动态计算权重。 + + :return: + """ + del self.sent_embeds diff --git a/fastNLP/embeddings/elmo_embedding.py b/fastNLP/embeddings/elmo_embedding.py new file mode 100644 index 00000000..53adfd62 --- /dev/null +++ b/fastNLP/embeddings/elmo_embedding.py @@ -0,0 +1,337 @@ + +import os + +import torch +import torch.nn as nn +import torch.nn.functional as F +import json +import codecs + +from ..core.vocabulary import Vocabulary +from ..io.file_utils import cached_path, _get_base_url, PRETRAINED_ELMO_MODEL_DIR +from ..modules.encoder._elmo import ElmobiLm, ConvTokenEmbedder +from .contextual_embedding import ContextualEmbedding + + +class ElmoEmbedding(ContextualEmbedding): + """ + 别名：:class:`fastNLP.embeddings.ElmoEmbedding` :class:`fastNLP.embeddings.elmo_embedding.ElmoEmbedding` + + 使用ELMo的embedding。初始化之后，只需要传入words就可以得到对应的embedding。当前支持的使用名称初始化的模型有以下的这些(待补充) + + Example:: + + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> # 使用不同层的concat的结果 + >>> embed = ElmoEmbedding(vocab, model_dir_or_name='en', layers='1,2', requires_grad=False) + >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) + >>> outputs = embed(words) + >>> outputs.size() + >>> # torch.Size([1, 5, 2048]) + + >>> # 使用不同层的weighted sum。 + >>> embed = ElmoEmbedding(vocab, model_dir_or_name='en', layers='mix', requires_grad=False) + >>> embed.set_mix_weights_requires_grad() # 使得weighted的权重是可以学习的，但ELMO的LSTM部分是不更新 + + :param vocab: 词表 + :param model_dir_or_name: 可以有两种方式调用预训练好的ELMo embedding：第一种是传入ELMo所在文件夹，该文件夹下面应该有两个文件， + 其中一个是以json为后缀的配置文件，另一个是以pkl为后缀的权重文件；第二种是传入ELMo版本的名称，将自动查看缓存中是否存在该模型， + 没有的话将自动下载并缓存。 + :param layers: str, 指定返回的层数(从0开始), 以,隔开不同的层。如果要返回第二层的结果'2', 返回后两层的结果'1,2'。不同的层的结果 + 按照这个顺序concat起来，默认为'2'。'mix'会使用可学习的权重结合不同层的表示(权重是否可训练与requires_grad保持一致， + 初始化权重对三层结果进行mean-pooling, 可以通过ElmoEmbedding.set_mix_weights_requires_grad()方法只将mix weights设置为可学习。) + :param requires_grad: bool, 该层是否需要gradient, 默认为False. + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 + :param cache_word_reprs: 可以选择对word的表示进行cache; 设置为True的话，将在初始化的时候为每个word生成对应的embedding， + 并删除character encoder，之后将直接使用cache的embedding。默认为False。 + """ + + def __init__(self, vocab: Vocabulary, model_dir_or_name: str = 'en', layers: str = '2', requires_grad: bool = False, + word_dropout=0.0, dropout=0.0, cache_word_reprs: bool = False): + super(ElmoEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + + # 根据model_dir_or_name检查是否存在并下载 + if model_dir_or_name.lower() in PRETRAINED_ELMO_MODEL_DIR: + PRETRAIN_URL = _get_base_url('elmo') + model_name = PRETRAINED_ELMO_MODEL_DIR[model_dir_or_name] + model_url = PRETRAIN_URL + model_name + model_dir = cached_path(model_url) + # 检查是否存在 + elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): + model_dir = model_dir_or_name + else: + raise ValueError(f"Cannot recognize {model_dir_or_name}.") + self.model = _ElmoModel(model_dir, vocab, cache_word_reprs=cache_word_reprs) + + if layers == 'mix': + self.layer_weights = nn.Parameter(torch.zeros(self.model.config['lstm']['n_layers'] + 1), + requires_grad=requires_grad) + self.gamma = nn.Parameter(torch.ones(1), requires_grad=requires_grad) + self._get_outputs = self._get_mixed_outputs + self._embed_size = self.model.config['lstm']['projection_dim'] * 2 + else: + layers = list(map(int, layers.split(','))) + assert len(layers) > 0, "Must choose one output" + for layer in layers: + assert 0 <= layer <= 2, "Layer index should be in range [0, 2]." + self.layers = layers + self._get_outputs = self._get_layer_outputs + self._embed_size = len(self.layers) * self.model.config['lstm']['projection_dim'] * 2 + + self.requires_grad = requires_grad + + def _get_mixed_outputs(self, outputs): + # outputs: num_layers x batch_size x max_len x hidden_size + # return: batch_size x max_len x hidden_size + weights = F.softmax(self.layer_weights + 1 / len(outputs), dim=0).to(outputs) + outputs = torch.einsum('l,lbij->bij', weights, outputs) + return self.gamma.to(outputs) * outputs + + def set_mix_weights_requires_grad(self, flag=True): + """ + 当初始化ElmoEmbedding时layers被设置为mix时，可以通过调用该方法设置mix weights是否可训练。如果layers不是mix，调用 + 该方法没有用。 + + :param bool flag: 混合不同层表示的结果是否可以训练。 + :return: + """ + if hasattr(self, 'layer_weights'): + self.layer_weights.requires_grad = flag + self.gamma.requires_grad = flag + + def _get_layer_outputs(self, outputs): + if len(self.layers) == 1: + outputs = outputs[self.layers[0]] + else: + outputs = torch.cat(tuple([*outputs[self.layers]]), dim=-1) + + return outputs + + def forward(self, words: torch.LongTensor): + """ + 计算words的elmo embedding表示。根据elmo文章中介绍的ELMO实际上是有2L+1层结果，但是为了让结果比较容易拆分，token的 + 被重复了一次，使得实际上layer=0的结果是[token_embedding;token_embedding], 而layer=1的结果是[forward_hiddens; + backward_hiddens]. + + :param words: batch_size x max_len + :return: torch.FloatTensor. batch_size x max_len x (512*len(self.layers)) + """ + words = self.drop_word(words) + outputs = self._get_sent_reprs(words) + if outputs is not None: + return self.dropout(outputs) + outputs = self.model(words) + outputs = self._get_outputs(outputs) + return self.dropout(outputs) + + def _delete_model_weights(self): + for name in ['layers', 'model', 'layer_weights', 'gamma']: + if hasattr(self, name): + delattr(self, name) + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + + :return: + """ + requires_grads = set([param.requires_grad for name, param in self.named_parameters() + if 'words_to_chars_embedding' not in name and 'words_to_words' not in name]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + if 'words_to_chars_embedding' in name or 'words_to_words' in name: # 这个不能加入到requires_grad中 + continue + param.requires_grad = value + + +class _ElmoModel(nn.Module): + """ + 该Module是ElmoEmbedding中进行所有的heavy lifting的地方。做的工作，包括 + (1) 根据配置，加载模型; + (2) 根据vocab，对模型中的embedding进行调整. 并将其正确初始化 + (3) 保存一个words与chars的对应转换，获取时自动进行相应的转换 + (4) 设计一个保存token的embedding，允许缓存word的表示。 + + """ + + def __init__(self, model_dir: str, vocab: Vocabulary = None, cache_word_reprs: bool = False): + super(_ElmoModel, self).__init__() + self.model_dir = model_dir + dir = os.walk(self.model_dir) + config_file = None + weight_file = None + config_count = 0 + weight_count = 0 + for path, dir_list, file_list in dir: + for file_name in file_list: + if file_name.__contains__(".json"): + config_file = file_name + config_count += 1 + elif file_name.__contains__(".pkl"): + weight_file = file_name + weight_count += 1 + if config_count > 1 or weight_count > 1: + raise Exception(f"Multiple config files(*.json) or weight files(*.hdf5) detected in {model_dir}.") + elif config_count == 0 or weight_count == 0: + raise Exception(f"No config file or weight file found in {model_dir}") + with open(os.path.join(model_dir, config_file), 'r') as config_f: + config = json.load(config_f) + self.weight_file = os.path.join(model_dir, weight_file) + self.config = config + + OOV_TAG = '' + PAD_TAG = '' + BOS_TAG = '' + EOS_TAG = '' + BOW_TAG = '' + EOW_TAG = '' + + # For the model trained with character-based word encoder. + char_lexicon = {} + with codecs.open(os.path.join(model_dir, 'char.dic'), 'r', encoding='utf-8') as fpi: + for line in fpi: + tokens = line.strip().split('\t') + if len(tokens) == 1: + tokens.insert(0, '\u3000') + token, i = tokens + char_lexicon[token] = int(i) + + # 做一些sanity check + for special_word in [PAD_TAG, OOV_TAG, BOW_TAG, EOW_TAG]: + assert special_word in char_lexicon, f"{special_word} not found in char.dic." + + # 从vocab中构建char_vocab + char_vocab = Vocabulary(unknown=OOV_TAG, padding=PAD_TAG) + # 需要保证与在里面 + char_vocab.add_word_lst([BOW_TAG, EOW_TAG, BOS_TAG, EOS_TAG]) + + for word, index in vocab: + char_vocab.add_word_lst(list(word)) + + self.bos_index, self.eos_index, self._pad_index = len(vocab), len(vocab) + 1, vocab.padding_idx + # 根据char_lexicon调整, 多设置一位，是预留给word padding的(该位置的char表示为全0表示) + char_emb_layer = nn.Embedding(len(char_vocab) + 1, int(config['char_cnn']['embedding']['dim']), + padding_idx=len(char_vocab)) + + # 读入预训练权重 这里的elmo_model 包含char_cnn和 lstm 的 state_dict + elmo_model = torch.load(os.path.join(self.model_dir, weight_file), map_location='cpu') + + char_embed_weights = elmo_model["char_cnn"]['char_emb_layer.weight'] + + found_char_count = 0 + for char, index in char_vocab: # 调整character embedding + if char in char_lexicon: + index_in_pre = char_lexicon.get(char) + found_char_count += 1 + else: + index_in_pre = char_lexicon[OOV_TAG] + char_emb_layer.weight.data[index] = char_embed_weights[index_in_pre] + + print(f"{found_char_count} out of {len(char_vocab)} characters were found in pretrained elmo embedding.") + # 生成words到chars的映射 + max_chars = config['char_cnn']['max_characters_per_token'] + + self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab) + 2, max_chars), + fill_value=len(char_vocab), + dtype=torch.long), + requires_grad=False) + for word, index in list(iter(vocab)) + [(BOS_TAG, len(vocab)), (EOS_TAG, len(vocab) + 1)]: + if len(word) + 2 > max_chars: + word = word[:max_chars - 2] + if index == self._pad_index: + continue + elif word == BOS_TAG or word == EOS_TAG: + char_ids = [char_vocab.to_index(BOW_TAG)] + [char_vocab.to_index(word)] + [ + char_vocab.to_index(EOW_TAG)] + char_ids += [char_vocab.to_index(PAD_TAG)] * (max_chars - len(char_ids)) + else: + char_ids = [char_vocab.to_index(BOW_TAG)] + [char_vocab.to_index(c) for c in word] + [ + char_vocab.to_index(EOW_TAG)] + char_ids += [char_vocab.to_index(PAD_TAG)] * (max_chars - len(char_ids)) + self.words_to_chars_embedding[index] = torch.LongTensor(char_ids) + + self.char_vocab = char_vocab + + self.token_embedder = ConvTokenEmbedder( + config, self.weight_file, None, char_emb_layer) + elmo_model["char_cnn"]['char_emb_layer.weight'] = char_emb_layer.weight + self.token_embedder.load_state_dict(elmo_model["char_cnn"]) + + self.output_dim = config['lstm']['projection_dim'] + + # lstm encoder + self.encoder = ElmobiLm(config) + self.encoder.load_state_dict(elmo_model["lstm"]) + + if cache_word_reprs: + if config['char_cnn']['embedding']['dim'] > 0: # 只有在使用了chars的情况下有用 + print("Start to generate cache word representations.") + batch_size = 320 + # bos eos + word_size = self.words_to_chars_embedding.size(0) + num_batches = word_size // batch_size + \ + int(word_size % batch_size != 0) + + self.cached_word_embedding = nn.Embedding(word_size, + config['lstm']['projection_dim']) + with torch.no_grad(): + for i in range(num_batches): + words = torch.arange(i * batch_size, + min((i + 1) * batch_size, word_size)).long() + chars = self.words_to_chars_embedding[words].unsqueeze(1) # batch_size x 1 x max_chars + word_reprs = self.token_embedder(words.unsqueeze(1), + chars).detach() # batch_size x 1 x config['encoder']['projection_dim'] + self.cached_word_embedding.weight.data[words] = word_reprs.squeeze(1) + + print("Finish generating cached word representations. Going to delete the character encoder.") + del self.token_embedder, self.words_to_chars_embedding + else: + print("There is no need to cache word representations, since no character information is used.") + + def forward(self, words): + """ + + :param words: batch_size x max_len + :return: num_layers x batch_size x max_len x hidden_size + """ + # 扩展, + batch_size, max_len = words.size() + expanded_words = words.new_zeros(batch_size, max_len + 2) # 因为pad一定为0， + seq_len = words.ne(self._pad_index).sum(dim=-1) + expanded_words[:, 1:-1] = words + expanded_words[:, 0].fill_(self.bos_index) + expanded_words[torch.arange(batch_size).to(words), seq_len + 1] = self.eos_index + seq_len = seq_len + 2 + zero_tensor = expanded_words.new_zeros(expanded_words.shape) + mask = (expanded_words == zero_tensor).unsqueeze(-1) + if hasattr(self, 'cached_word_embedding'): + token_embedding = self.cached_word_embedding(expanded_words) + else: + if hasattr(self, 'words_to_chars_embedding'): + chars = self.words_to_chars_embedding[expanded_words] + else: + chars = None + token_embedding = self.token_embedder(expanded_words, chars) # batch_size x max_len x embed_dim + + encoder_output = self.encoder(token_embedding, seq_len) + if encoder_output.size(2) < max_len + 2: + num_layers, _, output_len, hidden_size = encoder_output.size() + dummy_tensor = encoder_output.new_zeros(num_layers, batch_size, + max_len + 2 - output_len, hidden_size) + encoder_output = torch.cat((encoder_output, dummy_tensor), 2) + sz = encoder_output.size() # 2, batch_size, max_len, hidden_size + token_embedding = token_embedding.masked_fill(mask, 0) + token_embedding = torch.cat((token_embedding, token_embedding), dim=2).view(1, sz[1], sz[2], sz[3]) + encoder_output = torch.cat((token_embedding, encoder_output), dim=0) + + # 删除, . 这里没有精确地删除，但应该也不会影响最后的结果了。 + encoder_output = encoder_output[:, :, 1:-1] + return encoder_output diff --git a/fastNLP/embeddings/embedding.py b/fastNLP/embeddings/embedding.py new file mode 100644 index 00000000..9447c6ad --- /dev/null +++ b/fastNLP/embeddings/embedding.py @@ -0,0 +1,200 @@ +""" +该模块中的Embedding主要用于随机初始化的embedding(更推荐使用 :class:`fastNLP.embeddings.StaticEmbedding` )，或按照预训练权重初始化Embedding。 + +""" + + +import torch.nn as nn +from abc import abstractmethod +import torch + +from .utils import get_embeddings + + +class Embedding(nn.Module): + """ + 别名：:class:`fastNLP.embeddings.Embedding` :class:`fastNLP.embeddings.embedding.Embedding` + + 词向量嵌入，支持输入多种方式初始化. 可以通过self.num_embeddings获取词表大小; self.embedding_dim获取embedding的维度. + + Example:: + + >>> import numpy as np + >>> init_embed = (2000, 100) + >>> embed = Embedding(init_embed) # 随机初始化一个具有2000个词，每个词表示为100维的词向量 + >>> init_embed = np.zeros((2000, 100)) + >>> embed = Embedding(init_embed) # 使用numpy.ndarray的值作为初始化值初始化一个Embedding + + :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray init_embed: 支持传入Embedding的大小(传入tuple(int, int), + 第一个int为vocab_zie, 第二个int为embed_dim); 或传入Tensor, Embedding, numpy.ndarray等则直接使用该值初始化Embedding; + :param float word_dropout: 按照一定概率随机将word设置为unk_index，这样可以使得unk这个token得到足够的训练, 且会对网络有 + 一定的regularize的作用。设置该值时，必须同时设置unk_index + :param float dropout: 对Embedding的输出的dropout。 + :param int unk_index: drop word时替换为的index。fastNLP的Vocabulary的unk_index默认为1。 + """ + + def __init__(self, init_embed, word_dropout=0, dropout=0.0, unk_index=None): + + super(Embedding, self).__init__() + + self.embed = get_embeddings(init_embed) + + self.dropout = nn.Dropout(dropout) + if not isinstance(self.embed, TokenEmbedding): + if hasattr(self.embed, 'embed_size'): + self._embed_size = self.embed.embed_size + elif hasattr(self.embed, 'embedding_dim'): + self._embed_size = self.embed.embedding_dim + else: + self._embed_size = self.embed.weight.size(1) + if word_dropout>0 and not isinstance(unk_index, int): + raise ValueError("When drop word is set, you need to pass in the unk_index.") + else: + self._embed_size = self.embed.embed_size + unk_index = self.embed.get_word_vocab().unknown_idx + self.unk_index = unk_index + self.word_dropout = word_dropout + + def forward(self, words): + """ + :param torch.LongTensor words: [batch, seq_len] + :return: torch.Tensor : [batch, seq_len, embed_dim] + """ + if self.word_dropout>0 and self.training: + mask = torch.ones_like(words).float() * self.word_dropout + mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 + words = words.masked_fill(mask, self.unk_index) + words = self.embed(words) + return self.dropout(words) + + @property + def num_embedding(self)->int: + if isinstance(self.embed, nn.Embedding): + return self.embed.weight.size(0) + else: + return self.embed.num_embedding + + def __len__(self): + return len(self.embed) + + @property + def embed_size(self) -> int: + return self._embed_size + + @property + def embedding_dim(self) -> int: + return self._embed_size + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + :return: + """ + if not isinstance(self.embed, TokenEmbedding): + return self.embed.weight.requires_grad + else: + return self.embed.requires_grad + + @requires_grad.setter + def requires_grad(self, value): + if not isinstance(self.embed, TokenEmbedding): + self.embed.weight.requires_grad = value + else: + self.embed.requires_grad = value + + @property + def size(self): + if isinstance(self.embed, TokenEmbedding): + return self.embed.size + else: + return self.embed.weight.size() + + +class TokenEmbedding(nn.Module): + def __init__(self, vocab, word_dropout=0.0, dropout=0.0): + super(TokenEmbedding, self).__init__() + if vocab.rebuild: + vocab.build_vocab() + assert vocab.padding is not None, "Vocabulary must have a padding entry." + self._word_vocab = vocab + self._word_pad_index = vocab.padding_idx + if word_dropout>0: + assert vocab.unknown is not None, "Vocabulary must have unknown entry when you want to drop a word." + self.word_dropout = word_dropout + self._word_unk_index = vocab.unknown_idx + self.dropout_layer = nn.Dropout(dropout) + + def drop_word(self, words): + """ + 按照设定随机将words设置为unknown_index。 + + :param torch.LongTensor words: batch_size x max_len + :return: + """ + if self.word_dropout > 0 and self.training: + mask = torch.ones_like(words).float() * self.word_dropout + mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 + words = words.masked_fill(mask, self._word_unk_index) + return words + + def dropout(self, words): + """ + 对embedding后的word表示进行drop。 + + :param torch.FloatTensor words: batch_size x max_len x embed_size + :return: + """ + return self.dropout_layer(words) + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + :return: + """ + requires_grads = set([param.requires_grad for param in self.parameters()]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for param in self.parameters(): + param.requires_grad = value + + def __len__(self): + return len(self._word_vocab) + + @property + def embed_size(self) -> int: + return self._embed_size + + @property + def embedding_dim(self) -> int: + return self._embed_size + + @property + def num_embedding(self) -> int: + """ + 这个值可能会大于实际的embedding矩阵的大小。 + :return: + """ + return len(self._word_vocab) + + def get_word_vocab(self): + """ + 返回embedding的词典。 + + :return: Vocabulary + """ + return self._word_vocab + + @property + def size(self): + return torch.Size(self.num_embedding, self._embed_size) + + @abstractmethod + def forward(self, words): + raise NotImplementedError diff --git a/fastNLP/embeddings/stack_embedding.py b/fastNLP/embeddings/stack_embedding.py new file mode 100644 index 00000000..8091d598 --- /dev/null +++ b/fastNLP/embeddings/stack_embedding.py @@ -0,0 +1,94 @@ +from typing import List + +import torch +from torch import nn as nn + +from .embedding import TokenEmbedding + + +class StackEmbedding(TokenEmbedding): + """ + 别名：:class:`fastNLP.embeddings.StackEmbedding` :class:`fastNLP.embeddings.stack_embedding.StackEmbedding` + + 支持将多个embedding集合成一个embedding。 + + Example:: + + >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import StaticEmbedding + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) + >>> embed_2 = StaticEmbedding(vocab, model_dir_or_name='en-word2vec-300', requires_grad=True) + + :param embeds: 一个由若干个TokenEmbedding组成的list，要求每一个TokenEmbedding的词表都保持一致 + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。不同embedidng会在相同的位置 + 被设置为unknown。如果这里设置了dropout，则组成的embedding就不要再设置dropout了。 + :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 + + """ + def __init__(self, embeds: List[TokenEmbedding], word_dropout=0, dropout=0): + vocabs = [] + for embed in embeds: + if hasattr(embed, 'get_word_vocab'): + vocabs.append(embed.get_word_vocab()) + _vocab = vocabs[0] + for vocab in vocabs[1:]: + assert vocab == _vocab, "All embeddings in StackEmbedding should use the same word vocabulary." + + super(StackEmbedding, self).__init__(_vocab, word_dropout=word_dropout, dropout=dropout) + assert isinstance(embeds, list) + for embed in embeds: + assert isinstance(embed, TokenEmbedding), "Only TokenEmbedding type is supported." + self.embeds = nn.ModuleList(embeds) + self._embed_size = sum([embed.embed_size for embed in self.embeds]) + + def append(self, embed: TokenEmbedding): + """ + 添加一个embedding到结尾。 + :param embed: + :return: + """ + assert isinstance(embed, TokenEmbedding) + self.embeds.append(embed) + + def pop(self): + """ + 弹出最后一个embed + :return: + """ + return self.embeds.pop() + + @property + def embed_size(self): + return self._embed_size + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + :return: + """ + requires_grads = set([embed.requires_grad for embed in self.embeds()]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for embed in self.embeds(): + embed.requires_grad = value + + def forward(self, words): + """ + 得到多个embedding的结果，并把结果按照顺序concat起来。 + + :param words: batch_size x max_len + :return: 返回的shape和当前这个stack embedding中embedding的组成有关 + """ + outputs = [] + words = self.drop_word(words) + for embed in self.embeds: + outputs.append(embed(words)) + outputs = self.dropout(torch.cat(outputs, dim=-1)) + return outputs \ No newline at end of file diff --git a/fastNLP/embeddings/static_embedding.py b/fastNLP/embeddings/static_embedding.py new file mode 100644 index 00000000..b78e63e8 --- /dev/null +++ b/fastNLP/embeddings/static_embedding.py @@ -0,0 +1,255 @@ + +import os + +import torch +import torch.nn as nn +import numpy as np +import warnings + +from ..core.vocabulary import Vocabulary +from ..io.file_utils import PRETRAIN_STATIC_FILES, _get_base_url, cached_path +from .embedding import TokenEmbedding +from ..modules.utils import _get_file_name_base_on_postfix + +class StaticEmbedding(TokenEmbedding): + """ + 别名：:class:`fastNLP.embeddings.StaticEmbedding` :class:`fastNLP.embeddings.static_embedding.StaticEmbedding` + + StaticEmbedding组件. 给定预训练embedding的名称或路径，根据vocab从embedding中抽取相应的数据(只会将出现在vocab中的词抽取出来， + 如果没有找到，则会随机初始化一个值(但如果该word是被标记为no_create_entry的话，则不会单独创建一个值，而是会被指向unk的index))。 + 当前支持自动下载的预训练vector有以下的几种(待补充); + + Example:: + + >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) + >>> embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-50') + + >>> vocab = Vocabulary().add_word_lst(["The", 'the', "THE"]) + >>> embed = StaticEmbedding(vocab, model_dir_or_name="en-glove-50", lower=True) + >>> # "the", "The", "THE"它们共用一个vector，且将使用"the"在预训练词表中寻找它们的初始化表示。 + + >>> vocab = Vocabulary().add_word_lst(["The", "the", "THE"]) + >>> embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5, lower=True) + >>> words = torch.LongTensor([[vocab.to_index(word) for word in ["The", "the", "THE"]]]) + >>> embed(words) + >>> tensor([[[ 0.5773, 0.7251, -0.3104, 0.0777, 0.4849], + [ 0.5773, 0.7251, -0.3104, 0.0777, 0.4849], + [ 0.5773, 0.7251, -0.3104, 0.0777, 0.4849]]], + grad_fn=) # 每种word的输出是一致的。 + + :param vocab: Vocabulary. 若该项为None则会读取所有的embedding。 + :param model_dir_or_name: 可以有两种方式调用预训练好的static embedding：第一种是传入embedding文件夹(文件夹下应该只有一个 + 以.txt作为后缀的文件)或文件路径；第二种是传入embedding的名称，第二种情况将自动查看缓存中是否存在该模型，没有的话将自动下载。 + 如果输入为None则使用embedding_dim的维度随机初始化一个embedding。 + :param int embedding_dim: 随机初始化的embedding的维度，仅在model_dir_or_name为None时有效。 + :param bool requires_grad: 是否需要gradient. 默认为True + :param callable init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。调用该方法时传入一个tensor对 + :param bool lower: 是否将vocab中的词语小写后再和预训练的词表进行匹配。如果你的词表中包含大写的词语，或者就是需要单独 + 为大写的词语开辟一个vector表示，则将lower设置为False。 + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 + :param bool normalize: 是否对vector进行normalize，使得每个vector的norm为1。 + """ + def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en', embedding_dim=100, requires_grad: bool=True, + init_method=None, lower=False, dropout=0, word_dropout=0, normalize=False): + super(StaticEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + + # 得到cache_path + if model_dir_or_name is None: + assert embedding_dim>=1, "The dimension of embedding should be larger than 1." + embedding_dim = int(embedding_dim) + model_path = None + elif model_dir_or_name.lower() in PRETRAIN_STATIC_FILES: + PRETRAIN_URL = _get_base_url('static') + model_name = PRETRAIN_STATIC_FILES[model_dir_or_name] + model_url = PRETRAIN_URL + model_name + model_path = cached_path(model_url) + # 检查是否存在 + elif os.path.isfile(os.path.expanduser(os.path.abspath(model_dir_or_name))): + model_path = model_dir_or_name + elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): + model_path = _get_file_name_base_on_postfix(model_dir_or_name, '.txt') + else: + raise ValueError(f"Cannot recognize {model_dir_or_name}.") + + # 读取embedding + if lower: + lowered_vocab = Vocabulary(padding=vocab.padding, unknown=vocab.unknown) + for word, index in vocab: + if not vocab._is_word_no_create_entry(word): + lowered_vocab.add_word(word.lower()) # 先加入需要创建entry的 + for word in vocab._no_create_word.keys(): # 不需要创建entry的 + if word in vocab: + lowered_word = word.lower() + if lowered_word not in lowered_vocab.word_count: + lowered_vocab.add_word(lowered_word) + lowered_vocab._no_create_word[lowered_word] += 1 + print(f"All word in the vocab have been lowered. There are {len(vocab)} words, {len(lowered_vocab)} unique lowered " + f"words.") + if model_path: + embedding = self._load_with_vocab(model_path, vocab=lowered_vocab, init_method=init_method) + else: + embedding = self._randomly_init_embed(len(vocab), embedding_dim, init_method) + # 需要适配一下 + if not hasattr(self, 'words_to_words'): + self.words_to_words = torch.arange(len(lowered_vocab, )).long() + if lowered_vocab.unknown: + unknown_idx = lowered_vocab.unknown_idx + else: + unknown_idx = embedding.size(0) - 1 # 否则是最后一个为unknow + words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), + requires_grad=False) + for word, index in vocab: + if word not in lowered_vocab: + word = word.lower() + if lowered_vocab._is_word_no_create_entry(word): # 如果不需要创建entry,已经默认unknown了 + continue + words_to_words[index] = self.words_to_words[lowered_vocab.to_index(word)] + self.words_to_words = words_to_words + else: + if model_path: + embedding = self._load_with_vocab(model_path, vocab=vocab, init_method=init_method) + else: + embedding = self._randomly_init_embed(len(vocab), embedding_dim, init_method) + if normalize: + embedding /= (torch.norm(embedding, dim=1, keepdim=True) + 1e-12) + self.embedding = nn.Embedding(num_embeddings=embedding.shape[0], embedding_dim=embedding.shape[1], + padding_idx=vocab.padding_idx, + max_norm=None, norm_type=2, scale_grad_by_freq=False, + sparse=False, _weight=embedding) + self._embed_size = self.embedding.weight.size(1) + self.requires_grad = requires_grad + + def _randomly_init_embed(self, num_embedding, embedding_dim, init_embed=None): + """ + + :param int num_embedding: embedding的entry的数量 + :param int embedding_dim: embedding的维度大小 + :param callable init_embed: 初始化方法 + :return: torch.FloatTensor + """ + embed = torch.zeros(num_embedding, embedding_dim) + + if init_embed is None: + nn.init.uniform_(embed, -np.sqrt(3/embedding_dim), np.sqrt(3/embedding_dim)) + else: + init_embed(embed) + + return embed + + @property + def requires_grad(self): + """ + Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + + :return: + """ + requires_grads = set([param.requires_grad for name, param in self.named_parameters() + if 'words_to_words' not in name]) + if len(requires_grads) == 1: + return requires_grads.pop() + else: + return None + + @requires_grad.setter + def requires_grad(self, value): + for name, param in self.named_parameters(): + if 'words_to_words' in name: + continue + param.requires_grad = value + + def _load_with_vocab(self, embed_filepath, vocab, dtype=np.float32, padding='', unknown='', + error='ignore', init_method=None): + """ + 从embed_filepath这个预训练的词向量中抽取出vocab这个词表的词的embedding。EmbedLoader将自动判断embed_filepath是 + word2vec(第一行只有两个元素)还是glove格式的数据。 + + :param str embed_filepath: 预训练的embedding的路径。 + :param vocab: 词表 :class:`~fastNLP.Vocabulary` 类型，读取出现在vocab中的词的embedding。 + 没有出现在vocab中的词的embedding将通过找到的词的embedding的正态分布采样出来，以使得整个Embedding是同分布的。 + :param dtype: 读出的embedding的类型 + :param str padding: 词表中padding的token + :param str unknown: 词表中unknown的token + :param str error: `ignore` , `strict` ; 如果 `ignore` ，错误将自动跳过; 如果 `strict` , 错误将抛出。 + 这里主要可能出错的地方在于词表有空行或者词表出现了维度不一致。 + :param init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。默认使用torch.nn.init.zeros_ + :return torch.tensor: shape为 [len(vocab), dimension], dimension由pretrain的embedding决定。 + """ + assert isinstance(vocab, Vocabulary), "Only fastNLP.Vocabulary is supported." + if not os.path.exists(embed_filepath): + raise FileNotFoundError("`{}` does not exist.".format(embed_filepath)) + with open(embed_filepath, 'r', encoding='utf-8') as f: + line = f.readline().strip() + parts = line.split() + start_idx = 0 + if len(parts) == 2: + dim = int(parts[1]) + start_idx += 1 + else: + dim = len(parts) - 1 + f.seek(0) + matrix = {} + found_count = 0 + for idx, line in enumerate(f, start_idx): + try: + parts = line.strip().split() + word = ''.join(parts[:-dim]) + nums = parts[-dim:] + # 对齐unk与pad + if word == padding and vocab.padding is not None: + word = vocab.padding + elif word == unknown and vocab.unknown is not None: + word = vocab.unknown + if word in vocab: + index = vocab.to_index(word) + matrix[index] = torch.from_numpy(np.fromstring(' '.join(nums), sep=' ', dtype=dtype, count=dim)) + found_count += 1 + except Exception as e: + if error == 'ignore': + warnings.warn("Error occurred at the {} line.".format(idx)) + else: + print("Error occurred at the {} line.".format(idx)) + raise e + print("Found {} out of {} words in the pre-training embedding.".format(found_count, len(vocab))) + for word, index in vocab: + if index not in matrix and not vocab._is_word_no_create_entry(word): + if vocab.unknown_idx in matrix: # 如果有unkonwn，用unknown初始化 + matrix[index] = matrix[vocab.unknown_idx] + else: + matrix[index] = None + + vectors = self._randomly_init_embed(len(matrix), dim, init_method) + + if vocab._no_create_word_length>0: + if vocab.unknown is None: # 创建一个专门的unknown + unknown_idx = len(matrix) + vectors = torch.cat((vectors, torch.zeros(1, dim)), dim=0).contiguous() + else: + unknown_idx = vocab.unknown_idx + words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), + requires_grad=False) + for order, (index, vec) in enumerate(matrix.items()): + if vec is not None: + vectors[order] = vec + words_to_words[index] = order + self.words_to_words = words_to_words + else: + for index, vec in matrix.items(): + if vec is not None: + vectors[index] = vec + + return vectors + + def forward(self, words): + """ + 传入words的index + + :param words: torch.LongTensor, [batch_size, max_len] + :return: torch.FloatTensor, [batch_size, max_len, embed_size] + """ + if hasattr(self, 'words_to_words'): + words = self.words_to_words[words] + words = self.drop_word(words) + words = self.embedding(words) + words = self.dropout(words) + return words diff --git a/fastNLP/embeddings/utils.py b/fastNLP/embeddings/utils.py new file mode 100644 index 00000000..b79f563c --- /dev/null +++ b/fastNLP/embeddings/utils.py @@ -0,0 +1,51 @@ +import numpy as np +import torch +from torch import nn as nn + +from ..core.vocabulary import Vocabulary + +__all__ = ['get_embeddings'] + + +def _construct_char_vocab_from_vocab(vocab:Vocabulary, min_freq:int=1): + """ + 给定一个word的vocabulary生成character的vocabulary. + + :param vocab: 从vocab + :param min_freq: + :return: + """ + char_vocab = Vocabulary(min_freq=min_freq) + for word, index in vocab: + if not vocab._is_word_no_create_entry(word): + char_vocab.add_word_lst(list(word)) + return char_vocab + + +def get_embeddings(init_embed): + """ + 根据输入的init_embed返回Embedding对象。如果输入是tuple, 则随机初始化一个nn.Embedding; 如果输入是numpy.ndarray, 则按照ndarray + 的值将nn.Embedding初始化; 如果输入是torch.Tensor, 则按该值初始化nn.Embedding; 如果输入是fastNLP中的embedding将不做处理 + 返回原对象。 + + :param init_embed: 可以是 tuple:(num_embedings, embedding_dim), 即embedding的大小和每个词的维度;也可以传入 + nn.Embedding 对象, 此时就以传入的对象作为embedding; 传入np.ndarray也行，将使用传入的ndarray作为作为Embedding初始化; + 传入torch.Tensor, 将使用传入的值作为Embedding初始化。 + :return nn.Embedding embeddings: + """ + if isinstance(init_embed, tuple): + res = nn.Embedding( + num_embeddings=init_embed[0], embedding_dim=init_embed[1]) + nn.init.uniform_(res.weight.data, a=-np.sqrt(3/res.weight.data.size(1)), + b=np.sqrt(3/res.weight.data.size(1))) + elif isinstance(init_embed, nn.Module): + res = init_embed + elif isinstance(init_embed, torch.Tensor): + res = nn.Embedding.from_pretrained(init_embed, freeze=False) + elif isinstance(init_embed, np.ndarray): + init_embed = torch.tensor(init_embed, dtype=torch.float32) + res = nn.Embedding.from_pretrained(init_embed, freeze=False) + else: + raise TypeError( + 'invalid init_embed type: {}'.format((type(init_embed)))) + return res \ No newline at end of file diff --git a/fastNLP/io/__init__.py b/fastNLP/io/__init__.py index d1d1dc5d..cd0d3527 100644 --- a/fastNLP/io/__init__.py +++ b/fastNLP/io/__init__.py @@ -3,35 +3,40 @@ 1. 用于读入 embedding 的 :doc:`EmbedLoader ` 类, -2. 用于读入数据的 :doc:`DataSetLoader ` 类 +2. 用于读入不同格式数据的 :doc:`DataSetLoader ` 类 -3. 用于保存和载入模型的类, 参考 :doc:`/fastNLP.io.model_io` +3. 用于读入不同数据集并进行预处理的 :doc:`DataLoader ` 类 + +4. 用于保存和载入模型的类, 参考 :doc:`model_io文档` 这些类的使用方法如下: """ __all__ = [ 'EmbedLoader', - 'DataBundle', - 'DataSetLoader', - 'CSVLoader', 'JsonLoader', - - 'ModelLoader', - 'ModelSaver', + + 'DataBundle', + 'DataSetLoader', 'ConllLoader', 'Conll2003Loader', + 'IMDBLoader', 'MatchingLoader', - 'PeopleDailyCorpusLoader', 'SNLILoader', - 'SSTLoader', - 'SST2Loader', 'MNLILoader', + 'MTL16Loader', + 'PeopleDailyCorpusLoader', 'QNLILoader', 'QuoraLoader', 'RTELoader', + 'SSTLoader', + 'SST2Loader', + 'YelpLoader', + + 'ModelLoader', + 'ModelSaver', ] from .embed_loader import EmbedLoader diff --git a/fastNLP/io/base_loader.py b/fastNLP/io/base_loader.py index 62793836..5d61c16a 100644 --- a/fastNLP/io/base_loader.py +++ b/fastNLP/io/base_loader.py @@ -111,7 +111,7 @@ def _uncompress(src, dst): class DataBundle: """ - 经过处理的数据信息，包括一系列数据集（比如：分开的训练集、验证集和测试集）及它们所用的词表和词嵌入。 + 经过处理的数据信息，包括一系列数据集（比如：分开的训练集、验证集和测试集）以及各个field对应的vocabulary。 :param vocabs: 从名称(字符串)到 :class:`~fastNLP.Vocabulary` 类型的dict :param datasets: 从名称(字符串)到 :class:`~fastNLP.DataSet` 类型的dict diff --git a/fastNLP/io/data_loader/__init__.py b/fastNLP/io/data_loader/__init__.py index d4777ff8..5d6b08b0 100644 --- a/fastNLP/io/data_loader/__init__.py +++ b/fastNLP/io/data_loader/__init__.py @@ -1,13 +1,14 @@ """ -用于读数据集的模块, 具体包括: +用于读数据集的模块, 可以读取文本分类、序列标注、Matching任务的数据集 -这些模块的使用方法如下: +这些模块的具体介绍如下，您可以通过阅读 :doc:`教程` 来进行了解。 """ __all__ = [ 'ConllLoader', 'Conll2003Loader', 'IMDBLoader', 'MatchingLoader', + 'SNLILoader', 'MNLILoader', 'MTL16Loader', 'PeopleDailyCorpusLoader', @@ -16,7 +17,6 @@ __all__ = [ 'RTELoader', 'SSTLoader', 'SST2Loader', - 'SNLILoader', 'YelpLoader', ] diff --git a/fastNLP/io/data_loader/conll.py b/fastNLP/io/data_loader/conll.py index 61f4f61b..9b2402a2 100644 --- a/fastNLP/io/data_loader/conll.py +++ b/fastNLP/io/data_loader/conll.py @@ -10,7 +10,7 @@ class ConllLoader(DataSetLoader): 别名：:class:`fastNLP.io.ConllLoader` :class:`fastNLP.io.data_loader.ConllLoader` 读取Conll格式的数据. 数据格式详见 http://conll.cemantix.org/2012/data.html. 数据中以"-DOCSTART-"开头的行将被忽略，因为 - 该符号在conll 2003中被用为文档分割符。 + 该符号在conll 2003中被用为文档分割符。 列号从0开始, 每列对应内容为:: @@ -58,7 +58,7 @@ class ConllLoader(DataSetLoader): class Conll2003Loader(ConllLoader): """ - 别名：:class:`fastNLP.io.Conll2003Loader` :class:`fastNLP.io.dataset_loader.Conll2003Loader` + 别名：:class:`fastNLP.io.Conll2003Loader` :class:`fastNLP.io.data_loader.Conll2003Loader` 读取Conll2003数据 diff --git a/fastNLP/io/data_loader/imdb.py b/fastNLP/io/data_loader/imdb.py index bf53c5be..d3636cde 100644 --- a/fastNLP/io/data_loader/imdb.py +++ b/fastNLP/io/data_loader/imdb.py @@ -13,9 +13,12 @@ from ..utils import get_tokenizer class IMDBLoader(DataSetLoader): """ + 别名：:class:`fastNLP.io.IMDBLoader` :class:`fastNLP.io.data_loader.IMDBLoader` + 读取IMDB数据集，DataSet包含以下fields: words: list(str), 需要分类的文本 + target: str, 文本的标签 """ diff --git a/fastNLP/io/data_loader/matching.py b/fastNLP/io/data_loader/matching.py index cecaee96..481b5056 100644 --- a/fastNLP/io/data_loader/matching.py +++ b/fastNLP/io/data_loader/matching.py @@ -6,7 +6,7 @@ from ...core.const import Const from ...core.vocabulary import Vocabulary from ..base_loader import DataBundle, DataSetLoader from ..file_utils import _get_base_url, cached_path, PRETRAINED_BERT_MODEL_DIR -from ...modules.encoder._bert import BertTokenizer +from ...modules.encoder.bert import BertTokenizer class MatchingLoader(DataSetLoader): diff --git a/fastNLP/io/data_loader/mnli.py b/fastNLP/io/data_loader/mnli.py index 5d857533..65863f3d 100644 --- a/fastNLP/io/data_loader/mnli.py +++ b/fastNLP/io/data_loader/mnli.py @@ -12,7 +12,9 @@ class MNLILoader(MatchingLoader, CSVLoader): 读取MNLI数据集，读取的DataSet包含fields:: words1: list(str)，第一句文本, premise + words2: list(str), 第二句文本, hypothesis + target: str, 真实标签 数据来源: diff --git a/fastNLP/io/data_loader/mtl.py b/fastNLP/io/data_loader/mtl.py index 940ece51..cbca413d 100644 --- a/fastNLP/io/data_loader/mtl.py +++ b/fastNLP/io/data_loader/mtl.py @@ -10,9 +10,12 @@ from ..utils import check_dataloader_paths class MTL16Loader(CSVLoader): """ + 别名：:class:`fastNLP.io.MTL16Loader` :class:`fastNLP.io.data_loader.MTL16Loader` + 读取MTL16数据集，DataSet包含以下fields: words: list(str), 需要分类的文本 + target: str, 文本的标签 数据来源：https://pan.baidu.com/s/1c2L6vdA diff --git a/fastNLP/io/data_loader/people_daily.py b/fastNLP/io/data_loader/people_daily.py index d8c55aef..5efadb7d 100644 --- a/fastNLP/io/data_loader/people_daily.py +++ b/fastNLP/io/data_loader/people_daily.py @@ -7,7 +7,7 @@ from ...core.const import Const class PeopleDailyCorpusLoader(DataSetLoader): """ - 别名：:class:`fastNLP.io.PeopleDailyCorpusLoader` :class:`fastNLP.io.dataset_loader.PeopleDailyCorpusLoader` + 别名：:class:`fastNLP.io.PeopleDailyCorpusLoader` :class:`fastNLP.io.data_loader.PeopleDailyCorpusLoader` 读取人民日报数据集 """ diff --git a/fastNLP/io/data_loader/qnli.py b/fastNLP/io/data_loader/qnli.py index ff6302b2..84b0f3d6 100644 --- a/fastNLP/io/data_loader/qnli.py +++ b/fastNLP/io/data_loader/qnli.py @@ -12,7 +12,9 @@ class QNLILoader(MatchingLoader, CSVLoader): 读取QNLI数据集，读取的DataSet包含fields:: words1: list(str)，第一句文本, premise + words2: list(str), 第二句文本, hypothesis + target: str, 真实标签 数据来源: diff --git a/fastNLP/io/data_loader/quora.py b/fastNLP/io/data_loader/quora.py index 12cc42ce..d0ee41ec 100644 --- a/fastNLP/io/data_loader/quora.py +++ b/fastNLP/io/data_loader/quora.py @@ -12,7 +12,9 @@ class QuoraLoader(MatchingLoader, CSVLoader): 读取MNLI数据集，读取的DataSet包含fields:: words1: list(str)，第一句文本, premise + words2: list(str), 第二句文本, hypothesis + target: str, 真实标签 数据来源: diff --git a/fastNLP/io/data_loader/rte.py b/fastNLP/io/data_loader/rte.py index c6c64ef8..f8c5e2fc 100644 --- a/fastNLP/io/data_loader/rte.py +++ b/fastNLP/io/data_loader/rte.py @@ -12,7 +12,9 @@ class RTELoader(MatchingLoader, CSVLoader): 读取RTE数据集，读取的DataSet包含fields:: words1: list(str)，第一句文本, premise + words2: list(str), 第二句文本, hypothesis + target: str, 真实标签 数据来源: diff --git a/fastNLP/io/data_loader/snli.py b/fastNLP/io/data_loader/snli.py index 8334fcfd..1db0ac5b 100644 --- a/fastNLP/io/data_loader/snli.py +++ b/fastNLP/io/data_loader/snli.py @@ -12,7 +12,9 @@ class SNLILoader(MatchingLoader, JsonLoader): 读取SNLI数据集，读取的DataSet包含fields:: words1: list(str)，第一句文本, premise + words2: list(str), 第二句文本, hypothesis + target: str, 真实标签 数据来源: https://nlp.stanford.edu/projects/snli/snli_1.0.zip diff --git a/fastNLP/io/data_loader/sst.py b/fastNLP/io/data_loader/sst.py index df46b47f..6c06a9ce 100644 --- a/fastNLP/io/data_loader/sst.py +++ b/fastNLP/io/data_loader/sst.py @@ -104,7 +104,9 @@ class SSTLoader(DataSetLoader): class SST2Loader(CSVLoader): """ - 数据来源"SST":'https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FSST-2.zip?alt=media&token=aabc5f6b-e466-44a2-b9b4-cf6337f84ac8', + 别名：:class:`fastNLP.io.SST2Loader` :class:`fastNLP.io.data_loader.SST2Loader` + + 数据来源 SST: https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FSST-2.zip?alt=media&token=aabc5f6b-e466-44a2-b9b4-cf6337f84ac8 """ def __init__(self): @@ -132,6 +134,7 @@ class SST2Loader(CSVLoader): info = DataBundle() for name, path in paths.items(): dataset = self.load(path) + dataset.apply_field(lambda words:words.copy(), field_name='words', new_field_name='raw_words') datasets[name] = dataset def wordtochar(words): diff --git a/fastNLP/io/data_loader/yelp.py b/fastNLP/io/data_loader/yelp.py index c287a90c..333fcab0 100644 --- a/fastNLP/io/data_loader/yelp.py +++ b/fastNLP/io/data_loader/yelp.py @@ -13,12 +13,17 @@ from ..utils import check_dataloader_paths, get_tokenizer class YelpLoader(DataSetLoader): """ + 别名：:class:`fastNLP.io.YelpLoader` :class:`fastNLP.io.data_loader.YelpLoader` 读取Yelp_full/Yelp_polarity数据集, DataSet包含fields: + words: list(str), 需要分类的文本 + target: str, 文本的标签 + chars:list(str),未index的字符列表 数据集：yelp_full/yelp_polarity + :param fine_grained: 是否使用SST-5标准，若 ``False`` , 使用SST-2。Default: ``False`` :param lower: 是否需要自动转小写，默认为False。 """ diff --git a/fastNLP/io/file_utils.py b/fastNLP/io/file_utils.py index cb762eb7..4be1360b 100644 --- a/fastNLP/io/file_utils.py +++ b/fastNLP/io/file_utils.py @@ -11,7 +11,7 @@ import hashlib PRETRAINED_BERT_MODEL_DIR = { - 'en': 'bert-base-cased-f89bfe08.zip', + 'en': 'bert-large-cased-wwm.zip', 'en-base-uncased': 'bert-base-uncased-3413b23c.zip', 'en-base-cased': 'bert-base-cased-f89bfe08.zip', 'en-large-uncased': 'bert-large-uncased-20939f45.zip', @@ -24,14 +24,14 @@ PRETRAINED_BERT_MODEL_DIR = { 'cn': 'bert-base-chinese-29d0a84a.zip', 'cn-base': 'bert-base-chinese-29d0a84a.zip', - 'multilingual': 'bert-base-multilingual-cased-1bd364ee.zip', - 'multilingual-base-uncased': 'bert-base-multilingual-uncased-f8730fe4.zip', - 'multilingual-base-cased': 'bert-base-multilingual-cased-1bd364ee.zip', + 'multilingual': 'bert-base-multilingual-cased.zip', + 'multilingual-base-uncased': 'bert-base-multilingual-uncased.zip', + 'multilingual-base-cased': 'bert-base-multilingual-cased.zip', } PRETRAINED_ELMO_MODEL_DIR = { 'en': 'elmo_en-d39843fe.tar.gz', - 'cn': 'elmo_cn-5e9b34e2.tar.gz' + 'en-small': "elmo_en_Small.zip" } PRETRAIN_STATIC_FILES = { @@ -39,7 +39,7 @@ PRETRAIN_STATIC_FILES = { 'en-glove-840b-300': 'glove.840B.300d-cc1ad5e1.tar.gz', 'en-glove-6b-50': "glove.6B.50d-a6028c70.tar.gz", 'en-word2vec-300': "GoogleNews-vectors-negative300-be166d9d.tar.gz", - 'en-fasttext': "cc.en.300.vec-d53187b2.gz", + 'en-fasttext-wiki': "wiki-news-300d-1M.vec.zip", 'cn': "tencent_cn-dab24577.tar.gz", 'cn-fasttext': "cc.zh.300.vec-d68a9bcf.gz", } @@ -47,11 +47,15 @@ PRETRAIN_STATIC_FILES = { def cached_path(url_or_filename: str, cache_dir: Path=None) -> Path: """ - 给定一个url或者文件名(可以是具体的文件名，也可以是文件)，先在cache_dir下寻找该文件是否存在，如果不存在则去下载, 并 - 将文件放入到cache_dir中 + 给定一个url或者文件名(可以是具体的文件名，也可以是文件)，先在cache_dir下寻找该文件是否存在，如果不存在则去下载, 并 + 将文件放入到cache_dir中. + + :param url_or_filename: 文件的下载url或者文件路径 + :param cache_dir: 文件的缓存文件夹 + :return: """ if cache_dir is None: - dataset_cache = Path(get_defalt_path()) + dataset_cache = Path(get_default_cache_path()) else: dataset_cache = cache_dir @@ -75,7 +79,7 @@ def cached_path(url_or_filename: str, cache_dir: Path=None) -> Path: def get_filepath(filepath): """ - 如果filepath中只有一个文件，则直接返回对应的全路径 + 如果filepath中只有一个文件，则直接返回对应的全路径. :param filepath: :return: """ @@ -88,7 +92,7 @@ def get_filepath(filepath): return filepath -def get_defalt_path(): +def get_default_cache_path(): """ 获取默认的fastNLP存放路径, 如果将FASTNLP_CACHE_PATH设置在了环境变量中，将使用环境变量的值，使得不用每个用户都去下载。 @@ -96,11 +100,10 @@ def get_defalt_path(): """ if 'FASTNLP_CACHE_DIR' in os.environ: fastnlp_cache_dir = os.environ.get('FASTNLP_CACHE_DIR') - if os.path.exists(fastnlp_cache_dir): + if os.path.isdir(fastnlp_cache_dir): return fastnlp_cache_dir - raise RuntimeError("Some errors happens on cache directory.") - else: - raise RuntimeError("There function is not available right now.") + else: + raise NotADirectoryError(f"{os.environ['FASTNLP_CACHE_DIR']} is not a directory.") fastnlp_cache_dir = os.path.expanduser(os.path.join("~", ".fastNLP")) return fastnlp_cache_dir @@ -109,13 +112,19 @@ def _get_base_url(name): # 返回的URL结尾必须是/ if 'FASTNLP_BASE_URL' in os.environ: fastnlp_base_url = os.environ['FASTNLP_BASE_URL'] - return fastnlp_base_url - raise RuntimeError("There function is not available right now.") + if fastnlp_base_url.endswith('/'): + return fastnlp_base_url + else: + return fastnlp_base_url + '/' + else: + # TODO 替换 + dbbrain_url = "http://dbcloud.irocn.cn:8989/api/public/dl/" + return dbbrain_url def split_filename_suffix(filepath): """ - 给定filepath返回对应的name和suffix + 给定filepath返回对应的name和suffix. 如果后缀是多个点，仅支持.tar.gz类型 :param filepath: :return: filename, suffix """ @@ -135,13 +144,6 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: filename = re.sub(r".+/", "", url) dir_name, suffix = split_filename_suffix(filename) - sep_index = dir_name[::-1].index('-') - if sep_index<0: - check_sum = None - else: - check_sum = dir_name[-sep_index+1:] - sep_index = len(dir_name) if sep_index==-1 else -sep_index-1 - dir_name = dir_name[:sep_index] # 寻找与它名字匹配的内容, 而不关心后缀 match_dir_name = match_file(dir_name, cache_dir) @@ -154,11 +156,11 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: return get_filepath(cache_path) # make HEAD request to check ETag TODO ETag可以用来判断资源是否已经更新了，之后需要加上 - response = requests.head(url, headers={"User-Agent": "fastNLP"}) - if response.status_code != 200: - raise IOError( - f"HEAD request failed for url {url} with status code {response.status_code}." - ) + # response = requests.head(url, headers={"User-Agent": "fastNLP"}) + # if response.status_code != 200: + # raise IOError( + # f"HEAD request failed for url {url} with status code {response.status_code}." + # ) # add ETag to filename if it exists # etag = response.headers.get("ETag") @@ -174,17 +176,11 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: content_length = req.headers.get("Content-Length") total = int(content_length) if content_length is not None else None progress = tqdm(unit="B", total=total) - sha256 = hashlib.sha256() with open(temp_filename, "wb") as temp_file: for chunk in req.iter_content(chunk_size=1024): if chunk: # filter out keep-alive new chunks progress.update(len(chunk)) temp_file.write(chunk) - sha256.update(chunk) - # check sum - digit = sha256.hexdigest()[:8] - if not check_sum: - assert digit == check_sum, "File corrupted when download." progress.close() print(f"Finish download from {url}.") @@ -193,7 +189,7 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: if suffix in ('.zip', '.tar.gz'): uncompress_temp_dir = tempfile.mkdtemp() delete_temp_dir = uncompress_temp_dir - print(f"Start to uncompress file to {uncompress_temp_dir}.") + print(f"Start to uncompress file to {uncompress_temp_dir}") if suffix == '.zip': unzip_file(Path(temp_filename), Path(uncompress_temp_dir)) else: @@ -211,7 +207,7 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: success = False try: # 复制到指定的位置 - print(f"Copy file to {cache_path}.") + print(f"Copy file to {cache_path}") if os.path.isdir(uncompress_temp_dir): for filename in os.listdir(uncompress_temp_dir): shutil.copyfile(os.path.join(uncompress_temp_dir, filename), cache_path/filename) @@ -252,7 +248,7 @@ def untar_gz_file(file:Path, to:Path): tar.extractall(to) -def match_file(dir_name: str, cache_dir: str) -> str: +def match_file(dir_name: str, cache_dir: Path) -> str: """ 匹配的原则是，在cache_dir下的文件: (1) 与dir_name完全一致; (2) 除了后缀以外和dir_name完全一致。 如果找到了两个匹配的结果将报错. 如果找到了则返回匹配的文件的名称; 没有找到返回空字符串 @@ -273,27 +269,3 @@ def match_file(dir_name: str, cache_dir: str) -> str: else: raise RuntimeError(f"Duplicate matched files:{matched_filenames}, this should be caused by a bug.") - -if __name__ == '__main__': - cache_dir = Path('caches') - cache_dir = None - # 需要对cache_dir进行测试 - base_url = 'http://0.0.0.0:8888/file/download' - # if True: - # for filename in os.listdir(cache_dir): - # if os.path.isdir(os.path.join(cache_dir, filename)): - # shutil.rmtree(os.path.join(cache_dir, filename)) - # else: - # os.remove(os.path.join(cache_dir, filename)) - # 1. 测试.txt文件 - print(cached_path(base_url + '/{}'.format('txt_test-bcb4fe65.txt'), cache_dir)) - # 2. 测试.zip文件(只有一个文件) - print(cached_path(base_url + '/{}'.format('zip_test-40966d39.zip'), cache_dir)) - # 3. 测试.zip文件(有多个文件) - print(cached_path(base_url + '/{}'.format('zip_pack_test-70c0b20d.zip'), cache_dir)) - # 4. 测试.tar.gz文件 - print(cached_path(base_url + '/{}'.format('tar_gz_test-3e2679cf.tar.gz'), cache_dir)) - # 5. 测试.tar.gz多个文件 - print(cached_path(base_url + '/{}'.format('tar_gz_pack_test-08dfdccd.tar.gz'), cache_dir)) - - # 6. 测试.pkl文件 diff --git a/fastNLP/models/bert.py b/fastNLP/models/bert.py index fb186ce4..adecab60 100644 --- a/fastNLP/models/bert.py +++ b/fastNLP/models/bert.py @@ -8,7 +8,7 @@ from torch import nn from .base_model import BaseModel from ..core.const import Const from ..modules.encoder import BertModel -from ..modules.encoder._bert import BertConfig +from ..modules.encoder.bert import BertConfig class BertForSequenceClassification(BaseModel): diff --git a/fastNLP/models/biaffine_parser.py b/fastNLP/models/biaffine_parser.py index 8533e7af..29487864 100644 --- a/fastNLP/models/biaffine_parser.py +++ b/fastNLP/models/biaffine_parser.py @@ -20,7 +20,7 @@ from ..modules.dropout import TimestepDropout from ..modules.encoder.transformer import TransformerEncoder from ..modules.encoder.variational_rnn import VarLSTM from ..modules.utils import initial_parameter -from ..modules.utils import get_embeddings +from ..embeddings.utils import get_embeddings from .base_model import BaseModel from ..core.utils import seq_len_to_mask @@ -130,6 +130,8 @@ def _find_cycle(vertices, edges): class GraphParser(BaseModel): """ + 别名：:class:`fastNLP.models.GraphParser` :class:`fastNLP.models.baffine_parser.GraphParser` + 基于图的parser base class, 支持贪婪解码和最大生成树解码 """ diff --git a/fastNLP/models/cnn_text_classification.py b/fastNLP/models/cnn_text_classification.py index 081dd510..e00a0697 100644 --- a/fastNLP/models/cnn_text_classification.py +++ b/fastNLP/models/cnn_text_classification.py @@ -6,8 +6,9 @@ import torch import torch.nn as nn from ..core.const import Const as C +from ..core.utils import seq_len_to_mask from ..modules import encoder -from fastNLP import seq_len_to_mask +from ..embeddings import embedding class CNNText(torch.nn.Module): @@ -24,23 +25,23 @@ class CNNText(torch.nn.Module): :param int,tuple(int) kernel_sizes: 输出channel的kernel大小。 :param float dropout: Dropout的大小 """ - + def __init__(self, init_embed, num_classes, kernel_nums=(30, 40, 50), kernel_sizes=(1, 3, 5), dropout=0.5): super(CNNText, self).__init__() - + # no support for pre-trained embedding currently - self.embed = encoder.Embedding(init_embed) + self.embed = embedding.Embedding(init_embed) self.conv_pool = encoder.ConvMaxpool( in_channels=self.embed.embedding_dim, out_channels=kernel_nums, kernel_sizes=kernel_sizes) self.dropout = nn.Dropout(dropout) self.fc = nn.Linear(sum(kernel_nums), num_classes) - + def forward(self, words, seq_len=None): """ @@ -57,7 +58,7 @@ class CNNText(torch.nn.Module): x = self.dropout(x) x = self.fc(x) # [N,C] -> [N, N_class] return {C.OUTPUT: x} - + def predict(self, words, seq_len=None): """ :param torch.LongTensor words: [batch_size, seq_len]，句子中word的index diff --git a/fastNLP/models/enas_trainer.py b/fastNLP/models/enas_trainer.py index ef596b03..7abcc45f 100644 --- a/fastNLP/models/enas_trainer.py +++ b/fastNLP/models/enas_trainer.py @@ -14,7 +14,7 @@ except: from ..core.utils import _pseudo_tqdm as tqdm from ..core.trainer import Trainer -from ..core.batch import Batch +from ..core.batch import DataSetIter from ..core.callback import CallbackManager, CallbackException from ..core.dataset import DataSet from ..core.utils import _move_dict_value_to_device @@ -124,8 +124,8 @@ class ENASTrainer(Trainer): len(self.train_data) % self.batch_size != 0)) * self.n_epochs with inner_tqdm(total=total_steps, postfix='loss:{0:<6.5f}', leave=False, dynamic_ncols=True) as pbar: avg_loss = 0 - data_iterator = Batch(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) + data_iterator = DataSetIter(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, + prefetch=self.prefetch) for epoch in range(1, self.n_epochs + 1): pbar.set_description_str(desc="Epoch {}/{}".format(epoch, self.n_epochs)) last_stage = (epoch > self.n_epochs + 1 - self.final_epochs) @@ -209,8 +209,8 @@ class ENASTrainer(Trainer): total_loss = 0 train_idx = 0 avg_loss = 0 - data_iterator = Batch(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) + data_iterator = DataSetIter(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, + prefetch=self.prefetch) for batch_x, batch_y in data_iterator: _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) @@ -262,8 +262,8 @@ class ENASTrainer(Trainer): if not isinstance(entropies, np.ndarray): entropies = entropies.data.cpu().numpy() - data_iterator = Batch(self.dev_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) + data_iterator = DataSetIter(self.dev_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, + prefetch=self.prefetch) for inputs, targets in data_iterator: valid_loss, hidden, _ = self.get_loss(inputs, targets, hidden, dag) diff --git a/fastNLP/models/sequence_labeling.py b/fastNLP/models/sequence_labeling.py index 8e6a5db1..4bf3f95f 100644 --- a/fastNLP/models/sequence_labeling.py +++ b/fastNLP/models/sequence_labeling.py @@ -1,19 +1,82 @@ """ - 本模块实现了两种序列标注模型 + 本模块实现了几种序列标注模型 """ __all__ = [ "SeqLabeling", - "AdvSeqLabel" + "AdvSeqLabel", + # "BiLSTMCRF" ] import torch import torch.nn as nn +import torch.nn.functional as F from .base_model import BaseModel +from ..embeddings import embedding from ..modules import decoder, encoder from ..modules.decoder.crf import allowed_transitions from ..core.utils import seq_len_to_mask from ..core.const import Const as C +from ..modules import LSTM +from ..embeddings import get_embeddings +from ..modules import ConditionalRandomField + + +class BiLSTMCRF(BaseModel): + """ + 结构为BiLSTM + FC + Dropout + CRF. + + .. todo:: + 继续补充文档 + + :param embed: tuple: + :param num_classes: + :param num_layers: + :param hidden_size: + :param dropout: + :param target_vocab: + :param encoding_type: + """ + def __init__(self, embed, num_classes, num_layers=1, hidden_size=100, dropout=0.5, + target_vocab=None, encoding_type=None): + super().__init__() + self.embed = get_embeddings(embed) + + if num_layers>1: + self.lstm = LSTM(embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, + batch_first=True, dropout=dropout) + else: + self.lstm = LSTM(embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, + batch_first=True) + + self.dropout = nn.Dropout(dropout) + self.fc = nn.Linear(hidden_size, num_classes) + + trans = None + if target_vocab is not None and encoding_type is not None: + trans = allowed_transitions(target_vocab.idx2word, encoding_type=encoding_type, include_start_end=True) + + self.crf = ConditionalRandomField(num_classes, include_start_end_trans=True, allowed_transitions=trans) + + def _forward(self, words, seq_len=None, target=None): + words = self.embed(words) + feats = self.lstm(words, seq_len=seq_len) + feats = self.fc(feats) + feats = self.dropout(feats) + logits = F.log_softmax(feats, dim=-1) + mask = seq_len_to_mask(seq_len) + if target is None: + pred, _ = self.crf.viterbi_decode(logits, mask) + return {C.OUTPUT:pred} + else: + loss = self.crf(logits, target, mask).mean() + return {C.LOSS:loss} + + def forward(self, words, seq_len, target): + return self._forward(words, seq_len, target) + + def predict(self, words, seq_len): + return self._forward(words, seq_len) class SeqLabeling(BaseModel): @@ -32,10 +95,10 @@ class SeqLabeling(BaseModel): def __init__(self, init_embed, hidden_size, num_classes): super(SeqLabeling, self).__init__() - self.Embedding = encoder.embedding.Embedding(init_embed) - self.Rnn = encoder.lstm.LSTM(self.Embedding.embedding_dim, hidden_size) + self.Embedding = embedding.Embedding(init_embed) + self.Rnn = encoder.LSTM(self.Embedding.embedding_dim, hidden_size) self.Linear = nn.Linear(hidden_size, num_classes) - self.Crf = decoder.crf.ConditionalRandomField(num_classes) + self.Crf = decoder.ConditionalRandomField(num_classes) self.mask = None def forward(self, words, seq_len, target): @@ -129,7 +192,7 @@ class AdvSeqLabel(nn.Module): super().__init__() - self.Embedding = encoder.embedding.Embedding(init_embed) + self.Embedding = embedding.Embedding(init_embed) self.norm1 = torch.nn.LayerNorm(self.Embedding.embedding_dim) self.Rnn = encoder.LSTM(input_size=self.Embedding.embedding_dim, hidden_size=hidden_size, num_layers=2, dropout=dropout, diff --git a/fastNLP/models/snli.py b/fastNLP/models/snli.py index d12524cc..3be942e8 100644 --- a/fastNLP/models/snli.py +++ b/fastNLP/models/snli.py @@ -8,29 +8,34 @@ import torch.nn.functional as F from torch.nn import CrossEntropyLoss -from fastNLP.models import BaseModel -from fastNLP.modules.encoder.embedding import TokenEmbedding -from fastNLP.modules.encoder.lstm import LSTM -from fastNLP.core.const import Const -from fastNLP.core.utils import seq_len_to_mask +from .base_model import BaseModel +from ..embeddings.embedding import TokenEmbedding, Embedding +from ..core.const import Const +from ..core.utils import seq_len_to_mask class ESIM(BaseModel): - """ESIM model的一个PyTorch实现 + """ + 别名：:class:`fastNLP.models.ESIM` :class:`fastNLP.models.snli.ESIM` + + ESIM model的一个PyTorch实现 论文参见： https://arxiv.org/pdf/1609.06038.pdf - :param fastNLP.TokenEmbedding init_embedding: 初始化的TokenEmbedding + :param init_embedding: 初始化的Embedding :param int hidden_size: 隐藏层大小，默认值为Embedding的维度 :param int num_labels: 目标标签种类数量，默认值为3 :param float dropout_rate: dropout的比率，默认值为0.3 :param float dropout_embed: 对Embedding的dropout比率，默认值为0.1 """ - def __init__(self, init_embedding: TokenEmbedding, hidden_size=None, num_labels=3, dropout_rate=0.3, + def __init__(self, init_embedding, hidden_size=None, num_labels=3, dropout_rate=0.3, dropout_embed=0.1): super(ESIM, self).__init__() - self.embedding = init_embedding + if isinstance(init_embedding, TokenEmbedding) or isinstance(init_embedding, Embedding): + self.embedding = init_embedding + else: + self.embedding = Embedding(init_embedding) self.dropout_embed = EmbedDropout(p=dropout_embed) if hidden_size is None: hidden_size = self.embedding.embed_size diff --git a/fastNLP/models/star_transformer.py b/fastNLP/models/star_transformer.py index bb91a5b6..b95d1c25 100644 --- a/fastNLP/models/star_transformer.py +++ b/fastNLP/models/star_transformer.py @@ -13,7 +13,7 @@ from torch import nn from ..modules.encoder.star_transformer import StarTransformer from ..core.utils import seq_len_to_mask -from ..modules.utils import get_embeddings +from ..embeddings.utils import get_embeddings from ..core.const import Const @@ -34,7 +34,7 @@ class StarTransEnc(nn.Module): :param emb_dropout: 词嵌入的dropout概率. :param dropout: 模型除词嵌入外的dropout概率. """ - + def __init__(self, init_embed, hidden_size, num_layers, @@ -54,7 +54,7 @@ class StarTransEnc(nn.Module): head_dim=head_dim, dropout=dropout, max_len=max_len) - + def forward(self, x, mask): """ :param FloatTensor x: [batch, length, hidden] 输入的序列 @@ -79,7 +79,7 @@ class _Cls(nn.Module): nn.Dropout(dropout), nn.Linear(hid_dim, num_cls), ) - + def forward(self, x): h = self.fc(x) return h @@ -95,7 +95,7 @@ class _NLICls(nn.Module): nn.Dropout(dropout), nn.Linear(hid_dim, num_cls), ) - + def forward(self, x1, x2): x = torch.cat([x1, x2, torch.abs(x1 - x2), x1 * x2], 1) h = self.fc(x) @@ -121,7 +121,7 @@ class STSeqLabel(nn.Module): :param emb_dropout: 词嵌入的dropout概率. Default: 0.1 :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - + def __init__(self, init_embed, num_cls, hidden_size=300, num_layers=4, @@ -141,7 +141,7 @@ class STSeqLabel(nn.Module): emb_dropout=emb_dropout, dropout=dropout) self.cls = _Cls(hidden_size, num_cls, cls_hidden_size) - + def forward(self, words, seq_len): """ @@ -154,7 +154,7 @@ class STSeqLabel(nn.Module): output = self.cls(nodes) output = output.transpose(1, 2) # make hidden to be dim 1 return {Const.OUTPUT: output} # [bsz, n_cls, seq_len] - + def predict(self, words, seq_len): """ @@ -186,7 +186,7 @@ class STSeqCls(nn.Module): :param emb_dropout: 词嵌入的dropout概率. Default: 0.1 :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - + def __init__(self, init_embed, num_cls, hidden_size=300, num_layers=4, @@ -206,7 +206,7 @@ class STSeqCls(nn.Module): emb_dropout=emb_dropout, dropout=dropout) self.cls = _Cls(hidden_size, num_cls, cls_hidden_size, dropout=dropout) - + def forward(self, words, seq_len): """ @@ -219,7 +219,7 @@ class STSeqCls(nn.Module): y = 0.5 * (relay + nodes.max(1)[0]) output = self.cls(y) # [bsz, n_cls] return {Const.OUTPUT: output} - + def predict(self, words, seq_len): """ @@ -251,7 +251,7 @@ class STNLICls(nn.Module): :param emb_dropout: 词嵌入的dropout概率. Default: 0.1 :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - + def __init__(self, init_embed, num_cls, hidden_size=300, num_layers=4, @@ -271,7 +271,7 @@ class STNLICls(nn.Module): emb_dropout=emb_dropout, dropout=dropout) self.cls = _NLICls(hidden_size, num_cls, cls_hidden_size) - + def forward(self, words1, words2, seq_len1, seq_len2): """ @@ -283,16 +283,16 @@ class STNLICls(nn.Module): """ mask1 = seq_len_to_mask(seq_len1) mask2 = seq_len_to_mask(seq_len2) - + def enc(seq, mask): nodes, relay = self.enc(seq, mask) return 0.5 * (relay + nodes.max(1)[0]) - + y1 = enc(words1, mask1) y2 = enc(words2, mask2) output = self.cls(y1, y2) # [bsz, n_cls] return {Const.OUTPUT: output} - + def predict(self, words1, words2, seq_len1, seq_len2): """ diff --git a/fastNLP/modules/__init__.py b/fastNLP/modules/__init__.py index 2cd2216c..7959e454 100644 --- a/fastNLP/modules/__init__.py +++ b/fastNLP/modules/__init__.py @@ -1,46 +1,52 @@ """ -大部分用于的 NLP 任务神经网络都可以看做由编码 :mod:`~fastNLP.modules.encoder` 、 -解码 :mod:`~fastNLP.modules.decoder` 两种模块组成。 .. image:: figures/text_classification.png -:mod:`~fastNLP.modules` 中实现了 fastNLP 提供的诸多模块组件，可以帮助用户快速搭建自己所需的网络。 -两种模块的功能和常见组件如下: +大部分用于的 NLP 任务神经网络都可以看做由 :mod:`embedding` 、 :mod:`~fastNLP.modules.encoder` 、 +:mod:`~fastNLP.modules.decoder` 三种模块组成。 本模块中实现了 fastNLP 提供的诸多模块组件， +可以帮助用户快速搭建自己所需的网络。几种模块的功能和常见组件如下: + +.. csv-table:: + :header: "类型", "功能", "常见组件" + + "embedding", 参见 :doc:`/fastNLP.embeddings` , "Elmo, Bert" + "encoder", "将输入编码为具有表示能力的向量", "CNN, LSTM, Transformer" + "decoder", "将具有某种表示意义的向量解码为需要的输出形式 ", "MLP, CRF" + "其它", "配合其它组件使用的组件", "Dropout" -+-----------------------+-----------------------+-----------------------+ -| module type | functionality | example | -+=======================+=======================+=======================+ -| encoder | 将输入编码为具有具 | embedding, RNN, CNN, | -| | 有表示能力的向量 | transformer | -+-----------------------+-----------------------+-----------------------+ -| decoder | 将具有某种表示意义的 | MLP, CRF | -| | 向量解码为需要的输出 | | -| | 形式 | | -+-----------------------+-----------------------+-----------------------+ """ __all__ = [ # "BertModel", + "ConvolutionCharEncoder", "LSTMCharEncoder", + "ConvMaxpool", - "Embedding", + "LSTM", + "StarTransformer", + "TransformerEncoder", + "VarRNN", "VarLSTM", "VarGRU", - + "MaxPool", "MaxPoolWithMask", "AvgPool", + "AvgPoolWithMask", + "MultiHeadAttention", - + "MLP", "ConditionalRandomField", "viterbi_decode", "allowed_transitions", + + "TimestepDropout", ] from . import decoder @@ -48,4 +54,3 @@ from . import encoder from .decoder import * from .dropout import TimestepDropout from .encoder import * -from .utils import get_embeddings diff --git a/fastNLP/modules/decoder/crf.py b/fastNLP/modules/decoder/crf.py index c0717d6f..7c496868 100644 --- a/fastNLP/modules/decoder/crf.py +++ b/fastNLP/modules/decoder/crf.py @@ -11,7 +11,7 @@ from ..utils import initial_parameter def allowed_transitions(id2target, encoding_type='bio', include_start_end=False): """ - 别名：:class:`fastNLP.modules.allowed_transitions` :class:`fastNLP.modules.decoder.crf.allowed_transitions` + 别名：:class:`fastNLP.modules.allowed_transitions` :class:`fastNLP.modules.decoder.allowed_transitions` 给定一个id到label的映射表，返回所有可以跳转的(from_tag_id, to_tag_id)列表。 @@ -31,7 +31,7 @@ def allowed_transitions(id2target, encoding_type='bio', include_start_end=False) id_label_lst = list(id2target.items()) if include_start_end: id_label_lst += [(start_idx, 'start'), (end_idx, 'end')] - + def split_tag_label(from_label): from_label = from_label.lower() if from_label in ['start', 'end']: @@ -41,7 +41,7 @@ def allowed_transitions(id2target, encoding_type='bio', include_start_end=False) from_tag = from_label[:1] from_label = from_label[2:] return from_tag, from_label - + for from_id, from_label in id_label_lst: if from_label in ['', '']: continue @@ -93,7 +93,7 @@ def _is_transition_allowed(encoding_type, from_tag, from_label, to_tag, to_label return to_tag in ['end', 'b', 'o'] else: raise ValueError("Unexpect tag {}. Expect only 'B', 'I', 'O'.".format(from_tag)) - + elif encoding_type == 'bmes': """ 第一行是to_tag, 第一列是from_tag，y任意条件下可转，-只有在label相同时可转，n不可转 @@ -151,7 +151,7 @@ def _is_transition_allowed(encoding_type, from_tag, from_label, to_tag, to_label class ConditionalRandomField(nn.Module): """ - 别名：:class:`fastNLP.modules.ConditionalRandomField` :class:`fastNLP.modules.decoder.crf.ConditionalRandomField` + 别名：:class:`fastNLP.modules.ConditionalRandomField` :class:`fastNLP.modules.decoder.ConditionalRandomField` 条件随机场。 提供forward()以及viterbi_decode()两个方法，分别用于训练与inference。 @@ -163,21 +163,21 @@ class ConditionalRandomField(nn.Module): allowed_transitions()函数得到；如果为None，则所有跃迁均为合法 :param str initial_method: 初始化方法。见initial_parameter """ - + def __init__(self, num_tags, include_start_end_trans=False, allowed_transitions=None, initial_method=None): - + super(ConditionalRandomField, self).__init__() - + self.include_start_end_trans = include_start_end_trans self.num_tags = num_tags - + # the meaning of entry in this matrix is (from_tag_id, to_tag_id) score self.trans_m = nn.Parameter(torch.randn(num_tags, num_tags)) if self.include_start_end_trans: self.start_scores = nn.Parameter(torch.randn(num_tags)) self.end_scores = nn.Parameter(torch.randn(num_tags)) - + if allowed_transitions is None: constrain = torch.zeros(num_tags + 2, num_tags + 2) else: @@ -185,9 +185,9 @@ class ConditionalRandomField(nn.Module): for from_tag_id, to_tag_id in allowed_transitions: constrain[from_tag_id, to_tag_id] = 0 self._constrain = nn.Parameter(constrain, requires_grad=False) - + initial_parameter(self, initial_method) - + def _normalizer_likelihood(self, logits, mask): """Computes the (batch_size,) denominator term for the log-likelihood, which is the sum of the likelihoods across all possible state sequences. @@ -200,21 +200,21 @@ class ConditionalRandomField(nn.Module): alpha = logits[0] if self.include_start_end_trans: alpha = alpha + self.start_scores.view(1, -1) - + flip_mask = mask.eq(0) - + for i in range(1, seq_len): emit_score = logits[i].view(batch_size, 1, n_tags) trans_score = self.trans_m.view(1, n_tags, n_tags) tmp = alpha.view(batch_size, n_tags, 1) + emit_score + trans_score alpha = torch.logsumexp(tmp, 1).masked_fill(flip_mask[i].view(batch_size, 1), 0) + \ alpha.masked_fill(mask[i].byte().view(batch_size, 1), 0) - + if self.include_start_end_trans: alpha = alpha + self.end_scores.view(1, -1) - + return torch.logsumexp(alpha, 1) - + def _gold_score(self, logits, tags, mask): """ Compute the score for the gold path. @@ -226,7 +226,7 @@ class ConditionalRandomField(nn.Module): seq_len, batch_size, _ = logits.size() batch_idx = torch.arange(batch_size, dtype=torch.long, device=logits.device) seq_idx = torch.arange(seq_len, dtype=torch.long, device=logits.device) - + # trans_socre [L-1, B] mask = mask.byte() flip_mask = mask.eq(0) @@ -243,7 +243,7 @@ class ConditionalRandomField(nn.Module): score = score + st_scores + ed_scores # return [B,] return score - + def forward(self, feats, tags, mask): """ 用于计算CRF的前向loss，返回值为一个batch_size的FloatTensor，可能需要mean()求得loss。 @@ -258,9 +258,9 @@ class ConditionalRandomField(nn.Module): mask = mask.transpose(0, 1).float() all_path_score = self._normalizer_likelihood(feats, mask) gold_path_score = self._gold_score(feats, tags, mask) - + return all_path_score - gold_path_score - + def viterbi_decode(self, logits, mask, unpad=False): """给定一个特征矩阵以及转移分数矩阵，计算出最佳的路径以及对应的分数 @@ -277,7 +277,7 @@ class ConditionalRandomField(nn.Module): batch_size, seq_len, n_tags = logits.size() logits = logits.transpose(0, 1).data # L, B, H mask = mask.transpose(0, 1).data.byte() # L, B - + # dp vpath = logits.new_zeros((seq_len, batch_size, n_tags), dtype=torch.long) vscore = logits[0] @@ -286,7 +286,7 @@ class ConditionalRandomField(nn.Module): if self.include_start_end_trans: transitions[n_tags, :n_tags] += self.start_scores.data transitions[:n_tags, n_tags + 1] += self.end_scores.data - + vscore += transitions[n_tags, :n_tags] trans_score = transitions[:n_tags, :n_tags].view(1, n_tags, n_tags).data for i in range(1, seq_len): @@ -297,17 +297,17 @@ class ConditionalRandomField(nn.Module): vpath[i] = best_dst vscore = best_score.masked_fill(mask[i].eq(0).view(batch_size, 1), 0) + \ vscore.masked_fill(mask[i].view(batch_size, 1), 0) - + if self.include_start_end_trans: vscore += transitions[:n_tags, n_tags + 1].view(1, -1) - + # backtrace batch_idx = torch.arange(batch_size, dtype=torch.long, device=logits.device) seq_idx = torch.arange(seq_len, dtype=torch.long, device=logits.device) lens = (mask.long().sum(0) - 1) # idxes [L, B], batched idx from seq_len-1 to 0 idxes = (lens.view(1, -1) - seq_idx.view(-1, 1)) % seq_len - + ans = logits.new_empty((seq_len, batch_size), dtype=torch.long) ans_score, last_tags = vscore.max(1) ans[idxes[0], batch_idx] = last_tags diff --git a/fastNLP/modules/decoder/mlp.py b/fastNLP/modules/decoder/mlp.py index 418b3a77..9d9d80f2 100644 --- a/fastNLP/modules/decoder/mlp.py +++ b/fastNLP/modules/decoder/mlp.py @@ -10,7 +10,7 @@ from ..utils import initial_parameter class MLP(nn.Module): """ - 别名：:class:`fastNLP.modules.MLP` :class:`fastNLP.modules.decoder.mlp.MLP` + 别名：:class:`fastNLP.modules.MLP` :class:`fastNLP.modules.decoder.MLP` 多层感知器 @@ -40,7 +40,7 @@ class MLP(nn.Module): >>> print(x) >>> print(y) """ - + def __init__(self, size_layer, activation='relu', output_activation=None, initial_method=None, dropout=0.0): super(MLP, self).__init__() self.hiddens = nn.ModuleList() @@ -51,9 +51,9 @@ class MLP(nn.Module): self.output = nn.Linear(size_layer[i - 1], size_layer[i]) else: self.hiddens.append(nn.Linear(size_layer[i - 1], size_layer[i])) - + self.dropout = nn.Dropout(p=dropout) - + actives = { 'relu': nn.ReLU(), 'tanh': nn.Tanh(), @@ -82,7 +82,7 @@ class MLP(nn.Module): else: raise ValueError("should set activation correctly: {}".format(activation)) initial_parameter(self, initial_method) - + def forward(self, x): """ :param torch.Tensor x: MLP接受的输入 diff --git a/fastNLP/modules/decoder/utils.py b/fastNLP/modules/decoder/utils.py index 249f3ff6..9e773336 100644 --- a/fastNLP/modules/decoder/utils.py +++ b/fastNLP/modules/decoder/utils.py @@ -6,7 +6,7 @@ import torch def viterbi_decode(logits, transitions, mask=None, unpad=False): r""" - 别名：:class:`fastNLP.modules.viterbi_decode` :class:`fastNLP.modules.decoder.utils.viterbi_decode` + 别名：:class:`fastNLP.modules.viterbi_decode` :class:`fastNLP.modules.decoder.viterbi_decode` 给定一个特征矩阵以及转移分数矩阵，计算出最佳的路径以及对应的分数 @@ -30,11 +30,11 @@ def viterbi_decode(logits, transitions, mask=None, unpad=False): mask = mask.transpose(0, 1).data.byte() # L, B else: mask = logits.new_ones((seq_len, batch_size), dtype=torch.uint8) - + # dp vpath = logits.new_zeros((seq_len, batch_size, n_tags), dtype=torch.long) vscore = logits[0] - + trans_score = transitions.view(1, n_tags, n_tags).data for i in range(1, seq_len): prev_score = vscore.view(batch_size, n_tags, 1) @@ -44,14 +44,14 @@ def viterbi_decode(logits, transitions, mask=None, unpad=False): vpath[i] = best_dst vscore = best_score.masked_fill(mask[i].eq(0).view(batch_size, 1), 0) + \ vscore.masked_fill(mask[i].view(batch_size, 1), 0) - + # backtrace batch_idx = torch.arange(batch_size, dtype=torch.long, device=logits.device) seq_idx = torch.arange(seq_len, dtype=torch.long, device=logits.device) lens = (mask.long().sum(0) - 1) # idxes [L, B], batched idx from seq_len-1 to 0 idxes = (lens.view(1, -1) - seq_idx.view(-1, 1)) % seq_len - + ans = logits.new_empty((seq_len, batch_size), dtype=torch.long) ans_score, last_tags = vscore.max(1) ans[idxes[0], batch_idx] = last_tags diff --git a/fastNLP/modules/dropout.py b/fastNLP/modules/dropout.py index 1363165c..0ea2a2d9 100644 --- a/fastNLP/modules/dropout.py +++ b/fastNLP/modules/dropout.py @@ -5,10 +5,8 @@ import torch class TimestepDropout(torch.nn.Dropout): """ - 别名：:class:`fastNLP.modules.TimestepDropout` - - 接受的参数shape为``[batch_size, num_timesteps, embedding_dim)]`` 使用同一个mask(shape为``(batch_size, embedding_dim)``) - 在每个timestamp上做dropout。 + 传入参数的shape为 ``(batch_size, num_timesteps, embedding_dim)`` + 使用同一个shape为 ``(batch_size, embedding_dim)`` 的mask在每个timestamp上做dropout。 """ def forward(self, x): diff --git a/fastNLP/modules/encoder/__init__.py b/fastNLP/modules/encoder/__init__.py index 7b5bc070..1e99a0fd 100644 --- a/fastNLP/modules/encoder/__init__.py +++ b/fastNLP/modules/encoder/__init__.py @@ -1,25 +1,17 @@ __all__ = [ # "BertModel", - + "ConvolutionCharEncoder", "LSTMCharEncoder", - + "ConvMaxpool", - - "Embedding", - "StaticEmbedding", - "ElmoEmbedding", - "BertEmbedding", - "StackEmbedding", - "LSTMCharEmbedding", - "CNNCharEmbedding", - + "LSTM", - + "StarTransformer", - + "TransformerEncoder", - + "VarRNN", "VarLSTM", "VarGRU", @@ -31,12 +23,10 @@ __all__ = [ "MultiHeadAttention", ] -from ._bert import BertModel -from .bert import BertWordPieceEncoder + +from .bert import BertModel from .char_encoder import ConvolutionCharEncoder, LSTMCharEncoder from .conv_maxpool import ConvMaxpool -from .embedding import Embedding, StaticEmbedding, ElmoEmbedding, BertEmbedding, \ - StackEmbedding, LSTMCharEmbedding, CNNCharEmbedding from .lstm import LSTM from .star_transformer import StarTransformer from .transformer import TransformerEncoder diff --git a/fastNLP/modules/encoder/_bert.py b/fastNLP/modules/encoder/_bert.py deleted file mode 100644 index 61a5d7d1..00000000 --- a/fastNLP/modules/encoder/_bert.py +++ /dev/null @@ -1,1069 +0,0 @@ - - - -""" -这个页面的代码很大程度上参考(复制粘贴)了https://github.com/huggingface/pytorch-pretrained-BERT的代码， 如果你发现该代码对你 - 有用，也请引用一下他们。 -""" - - - -from ...core.vocabulary import Vocabulary -import collections - -import unicodedata -import numpy as np -from itertools import chain -import copy -import json -import math -import os - -import torch -from torch import nn -import glob -import sys - -CONFIG_FILE = 'bert_config.json' - - -class BertConfig(object): - """Configuration class to store the configuration of a `BertModel`. - """ - def __init__(self, - vocab_size_or_config_json_file, - hidden_size=768, - num_hidden_layers=12, - num_attention_heads=12, - intermediate_size=3072, - hidden_act="gelu", - hidden_dropout_prob=0.1, - attention_probs_dropout_prob=0.1, - max_position_embeddings=512, - type_vocab_size=2, - initializer_range=0.02, - layer_norm_eps=1e-12): - """Constructs BertConfig. - - Args: - vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `BertModel`. - hidden_size: Size of the encoder layers and the pooler layer. - num_hidden_layers: Number of hidden layers in the Transformer encoder. - num_attention_heads: Number of attention heads for each attention layer in - the Transformer encoder. - intermediate_size: The size of the "intermediate" (i.e., feed-forward) - layer in the Transformer encoder. - hidden_act: The non-linear activation function (function or string) in the - encoder and pooler. If string, "gelu", "relu" and "swish" are supported. - hidden_dropout_prob: The dropout probabilitiy for all fully connected - layers in the embeddings, encoder, and pooler. - attention_probs_dropout_prob: The dropout ratio for the attention - probabilities. - max_position_embeddings: The maximum sequence length that this model might - ever be used with. Typically set this to something large just in case - (e.g., 512 or 1024 or 2048). - type_vocab_size: The vocabulary size of the `token_type_ids` passed into - `BertModel`. - initializer_range: The sttdev of the truncated_normal_initializer for - initializing all weight matrices. - layer_norm_eps: The epsilon used by LayerNorm. - """ - if isinstance(vocab_size_or_config_json_file, str) or (sys.version_info[0] == 2 - and isinstance(vocab_size_or_config_json_file, unicode)): - with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader: - json_config = json.loads(reader.read()) - for key, value in json_config.items(): - self.__dict__[key] = value - elif isinstance(vocab_size_or_config_json_file, int): - self.vocab_size = vocab_size_or_config_json_file - self.hidden_size = hidden_size - self.num_hidden_layers = num_hidden_layers - self.num_attention_heads = num_attention_heads - self.hidden_act = hidden_act - self.intermediate_size = intermediate_size - self.hidden_dropout_prob = hidden_dropout_prob - self.attention_probs_dropout_prob = attention_probs_dropout_prob - self.max_position_embeddings = max_position_embeddings - self.type_vocab_size = type_vocab_size - self.initializer_range = initializer_range - self.layer_norm_eps = layer_norm_eps - else: - raise ValueError("First argument must be either a vocabulary size (int)" - "or the path to a pretrained model config file (str)") - - @classmethod - def from_dict(cls, json_object): - """Constructs a `BertConfig` from a Python dictionary of parameters.""" - config = BertConfig(vocab_size_or_config_json_file=-1) - for key, value in json_object.items(): - config.__dict__[key] = value - return config - - @classmethod - def from_json_file(cls, json_file): - """Constructs a `BertConfig` from a json file of parameters.""" - with open(json_file, "r", encoding='utf-8') as reader: - text = reader.read() - return cls.from_dict(json.loads(text)) - - def __repr__(self): - return str(self.to_json_string()) - - def to_dict(self): - """Serializes this instance to a Python dictionary.""" - output = copy.deepcopy(self.__dict__) - return output - - def to_json_string(self): - """Serializes this instance to a JSON string.""" - return json.dumps(self.to_dict(), indent=2, sort_keys=True) + "\n" - - def to_json_file(self, json_file_path): - """ Save this instance to a json file.""" - with open(json_file_path, "w", encoding='utf-8') as writer: - writer.write(self.to_json_string()) - - -def gelu(x): - return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0))) - - -def swish(x): - return x * torch.sigmoid(x) - - -ACT2FN = {"gelu": gelu, "relu": torch.nn.functional.relu, "swish": swish} - - -class BertLayerNorm(nn.Module): - def __init__(self, hidden_size, eps=1e-12): - """Construct a layernorm module in the TF style (epsilon inside the square root). - """ - super(BertLayerNorm, self).__init__() - self.weight = nn.Parameter(torch.ones(hidden_size)) - self.bias = nn.Parameter(torch.zeros(hidden_size)) - self.variance_epsilon = eps - - def forward(self, x): - u = x.mean(-1, keepdim=True) - s = (x - u).pow(2).mean(-1, keepdim=True) - x = (x - u) / torch.sqrt(s + self.variance_epsilon) - return self.weight * x + self.bias - - -class BertEmbeddings(nn.Module): - """Construct the embeddings from word, position and token_type embeddings. - """ - def __init__(self, config): - super(BertEmbeddings, self).__init__() - self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=0) - self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size) - self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size) - - # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load - # any TensorFlow checkpoint file - self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - - def forward(self, input_ids, token_type_ids=None): - seq_length = input_ids.size(1) - position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device) - position_ids = position_ids.unsqueeze(0).expand_as(input_ids) - if token_type_ids is None: - token_type_ids = torch.zeros_like(input_ids) - - words_embeddings = self.word_embeddings(input_ids) - position_embeddings = self.position_embeddings(position_ids) - token_type_embeddings = self.token_type_embeddings(token_type_ids) - - embeddings = words_embeddings + position_embeddings + token_type_embeddings - embeddings = self.LayerNorm(embeddings) - embeddings = self.dropout(embeddings) - return embeddings - - -class BertSelfAttention(nn.Module): - def __init__(self, config): - super(BertSelfAttention, self).__init__() - if config.hidden_size % config.num_attention_heads != 0: - raise ValueError( - "The hidden size (%d) is not a multiple of the number of attention " - "heads (%d)" % (config.hidden_size, config.num_attention_heads)) - self.num_attention_heads = config.num_attention_heads - self.attention_head_size = int(config.hidden_size / config.num_attention_heads) - self.all_head_size = self.num_attention_heads * self.attention_head_size - - self.query = nn.Linear(config.hidden_size, self.all_head_size) - self.key = nn.Linear(config.hidden_size, self.all_head_size) - self.value = nn.Linear(config.hidden_size, self.all_head_size) - - self.dropout = nn.Dropout(config.attention_probs_dropout_prob) - - def transpose_for_scores(self, x): - new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size) - x = x.view(*new_x_shape) - return x.permute(0, 2, 1, 3) - - def forward(self, hidden_states, attention_mask): - mixed_query_layer = self.query(hidden_states) - mixed_key_layer = self.key(hidden_states) - mixed_value_layer = self.value(hidden_states) - - query_layer = self.transpose_for_scores(mixed_query_layer) - key_layer = self.transpose_for_scores(mixed_key_layer) - value_layer = self.transpose_for_scores(mixed_value_layer) - - # Take the dot product between "query" and "key" to get the raw attention scores. - attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2)) - attention_scores = attention_scores / math.sqrt(self.attention_head_size) - # Apply the attention mask is (precomputed for all layers in BertModel forward() function) - attention_scores = attention_scores + attention_mask - - # Normalize the attention scores to probabilities. - attention_probs = nn.Softmax(dim=-1)(attention_scores) - - # This is actually dropping out entire tokens to attend to, which might - # seem a bit unusual, but is taken from the original Transformer paper. - attention_probs = self.dropout(attention_probs) - - context_layer = torch.matmul(attention_probs, value_layer) - context_layer = context_layer.permute(0, 2, 1, 3).contiguous() - new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,) - context_layer = context_layer.view(*new_context_layer_shape) - return context_layer - - -class BertSelfOutput(nn.Module): - def __init__(self, config): - super(BertSelfOutput, self).__init__() - self.dense = nn.Linear(config.hidden_size, config.hidden_size) - self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - - def forward(self, hidden_states, input_tensor): - hidden_states = self.dense(hidden_states) - hidden_states = self.dropout(hidden_states) - hidden_states = self.LayerNorm(hidden_states + input_tensor) - return hidden_states - - -class BertAttention(nn.Module): - def __init__(self, config): - super(BertAttention, self).__init__() - self.self = BertSelfAttention(config) - self.output = BertSelfOutput(config) - - def forward(self, input_tensor, attention_mask): - self_output = self.self(input_tensor, attention_mask) - attention_output = self.output(self_output, input_tensor) - return attention_output - - -class BertIntermediate(nn.Module): - def __init__(self, config): - super(BertIntermediate, self).__init__() - self.dense = nn.Linear(config.hidden_size, config.intermediate_size) - if isinstance(config.hidden_act, str) or (sys.version_info[0] == 2 and isinstance(config.hidden_act, unicode)): - self.intermediate_act_fn = ACT2FN[config.hidden_act] - else: - self.intermediate_act_fn = config.hidden_act - - def forward(self, hidden_states): - hidden_states = self.dense(hidden_states) - hidden_states = self.intermediate_act_fn(hidden_states) - return hidden_states - - -class BertOutput(nn.Module): - def __init__(self, config): - super(BertOutput, self).__init__() - self.dense = nn.Linear(config.intermediate_size, config.hidden_size) - self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - - def forward(self, hidden_states, input_tensor): - hidden_states = self.dense(hidden_states) - hidden_states = self.dropout(hidden_states) - hidden_states = self.LayerNorm(hidden_states + input_tensor) - return hidden_states - - -class BertLayer(nn.Module): - def __init__(self, config): - super(BertLayer, self).__init__() - self.attention = BertAttention(config) - self.intermediate = BertIntermediate(config) - self.output = BertOutput(config) - - def forward(self, hidden_states, attention_mask): - attention_output = self.attention(hidden_states, attention_mask) - intermediate_output = self.intermediate(attention_output) - layer_output = self.output(intermediate_output, attention_output) - return layer_output - - -class BertEncoder(nn.Module): - def __init__(self, config): - super(BertEncoder, self).__init__() - layer = BertLayer(config) - self.layer = nn.ModuleList([copy.deepcopy(layer) for _ in range(config.num_hidden_layers)]) - - def forward(self, hidden_states, attention_mask, output_all_encoded_layers=True): - all_encoder_layers = [] - for layer_module in self.layer: - hidden_states = layer_module(hidden_states, attention_mask) - if output_all_encoded_layers: - all_encoder_layers.append(hidden_states) - if not output_all_encoded_layers: - all_encoder_layers.append(hidden_states) - return all_encoder_layers - - -class BertPooler(nn.Module): - def __init__(self, config): - super(BertPooler, self).__init__() - self.dense = nn.Linear(config.hidden_size, config.hidden_size) - self.activation = nn.Tanh() - - def forward(self, hidden_states): - # We "pool" the model by simply taking the hidden state corresponding - # to the first token. - first_token_tensor = hidden_states[:, 0] - pooled_output = self.dense(first_token_tensor) - pooled_output = self.activation(pooled_output) - return pooled_output - - -class BertModel(nn.Module): - """BERT(Bidirectional Embedding Representations from Transformers). - - 如果你想使用预训练好的权重矩阵，请在以下网址下载. - sources:: - - 'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin", - 'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-pytorch_model.bin", - 'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-pytorch_model.bin", - 'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-pytorch_model.bin", - 'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-pytorch_model.bin", - 'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-pytorch_model.bin", - 'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-pytorch_model.bin", - 'bert-base-german-cased': "https://int-deepset-models-bert.s3.eu-central-1.amazonaws.com/pytorch/bert-base-german-cased-pytorch_model.bin", - 'bert-large-uncased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-pytorch_model.bin", - 'bert-large-cased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-pytorch_model.bin", - 'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-pytorch_model.bin", - 'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-pytorch_model.bin", - 'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-pytorch_model.bin" - - - 用预训练权重矩阵来建立BERT模型:: - - model = BertModel.from_pretrained("path/to/weights/directory") - - 用随机初始化权重矩阵来建立BERT模型:: - - model = BertModel() - - :param int vocab_size: 词表大小，默认值为30522，为BERT English uncase版本的词表大小 - :param int hidden_size: 隐层大小，默认值为768，为BERT base的版本 - :param int num_hidden_layers: 隐藏层数，默认值为12，为BERT base的版本 - :param int num_attention_heads: 多头注意力头数，默认值为12，为BERT base的版本 - :param int intermediate_size: FFN隐藏层大小，默认值是3072，为BERT base的版本 - :param str hidden_act: FFN隐藏层激活函数，默认值为``gelu`` - :param float hidden_dropout_prob: FFN隐藏层dropout，默认值为0.1 - :param float attention_probs_dropout_prob: Attention层的dropout，默认值为0.1 - :param int max_position_embeddings: 最大的序列长度，默认值为512， - :param int type_vocab_size: 最大segment数量，默认值为2 - :param int initializer_range: 初始化权重范围，默认值为0.02 - """ - - def __init__(self, config, *inputs, **kwargs): - super(BertModel, self).__init__() - if not isinstance(config, BertConfig): - raise ValueError( - "Parameter config in `{}(config)` should be an instance of class `BertConfig`. " - "To create a model from a Google pretrained model use " - "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format( - self.__class__.__name__, self.__class__.__name__ - )) - super(BertModel, self).__init__() - self.config = config - self.hidden_size = self.config.hidden_size - self.embeddings = BertEmbeddings(config) - self.encoder = BertEncoder(config) - self.pooler = BertPooler(config) - self.apply(self.init_bert_weights) - - def init_bert_weights(self, module): - """ Initialize the weights. - """ - if isinstance(module, (nn.Linear, nn.Embedding)): - # Slightly different from the TF version which uses truncated_normal for initialization - # cf https://github.com/pytorch/pytorch/pull/5617 - module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) - elif isinstance(module, BertLayerNorm): - module.bias.data.zero_() - module.weight.data.fill_(1.0) - if isinstance(module, nn.Linear) and module.bias is not None: - module.bias.data.zero_() - - def forward(self, input_ids, token_type_ids=None, attention_mask=None, output_all_encoded_layers=True): - if attention_mask is None: - attention_mask = torch.ones_like(input_ids) - if token_type_ids is None: - token_type_ids = torch.zeros_like(input_ids) - - # We create a 3D attention mask from a 2D tensor mask. - # Sizes are [batch_size, 1, 1, to_seq_length] - # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length] - # this attention mask is more simple than the triangular masking of causal attention - # used in OpenAI GPT, we just need to prepare the broadcast dimension here. - extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2) - - # Since attention_mask is 1.0 for positions we want to attend and 0.0 for - # masked positions, this operation will create a tensor which is 0.0 for - # positions we want to attend and -10000.0 for masked positions. - # Since we are adding it to the raw scores before the softmax, this is - # effectively the same as removing these entirely. - extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility - extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0 - - embedding_output = self.embeddings(input_ids, token_type_ids) - encoded_layers = self.encoder(embedding_output, - extended_attention_mask, - output_all_encoded_layers=output_all_encoded_layers) - sequence_output = encoded_layers[-1] - pooled_output = self.pooler(sequence_output) - if not output_all_encoded_layers: - encoded_layers = encoded_layers[-1] - return encoded_layers, pooled_output - - @classmethod - def from_pretrained(cls, pretrained_model_dir, *inputs, **kwargs): - state_dict = kwargs.get('state_dict', None) - kwargs.pop('state_dict', None) - cache_dir = kwargs.get('cache_dir', None) - kwargs.pop('cache_dir', None) - from_tf = kwargs.get('from_tf', False) - kwargs.pop('from_tf', None) - # Load config - config_file = os.path.join(pretrained_model_dir, CONFIG_FILE) - config = BertConfig.from_json_file(config_file) - # logger.info("Model config {}".format(config)) - # Instantiate model. - model = cls(config, *inputs, **kwargs) - if state_dict is None: - files = glob.glob(os.path.join(pretrained_model_dir, '*.bin')) - if len(files)==0: - raise FileNotFoundError(f"There is no *.bin file in {pretrained_model_dir}") - elif len(files)>1: - raise FileExistsError(f"There are multiple *.bin files in {pretrained_model_dir}") - weights_path = files[0] - state_dict = torch.load(weights_path, map_location='cpu') - - old_keys = [] - new_keys = [] - for key in state_dict.keys(): - new_key = None - if 'gamma' in key: - new_key = key.replace('gamma', 'weight') - if 'beta' in key: - new_key = key.replace('beta', 'bias') - if new_key: - old_keys.append(key) - new_keys.append(new_key) - for old_key, new_key in zip(old_keys, new_keys): - state_dict[new_key] = state_dict.pop(old_key) - - missing_keys = [] - unexpected_keys = [] - error_msgs = [] - # copy state_dict so _load_from_state_dict can modify it - metadata = getattr(state_dict, '_metadata', None) - state_dict = state_dict.copy() - if metadata is not None: - state_dict._metadata = metadata - - def load(module, prefix=''): - local_metadata = {} if metadata is None else metadata.get(prefix[:-1], {}) - module._load_from_state_dict( - state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs) - for name, child in module._modules.items(): - if child is not None: - load(child, prefix + name + '.') - - load(model, prefix='' if hasattr(model, 'bert') else 'bert.') - if len(missing_keys) > 0: - print("Weights of {} not initialized from pretrained model: {}".format( - model.__class__.__name__, missing_keys)) - if len(unexpected_keys) > 0: - print("Weights from pretrained model not used in {}: {}".format( - model.__class__.__name__, unexpected_keys)) - return model - - -def whitespace_tokenize(text): - """Runs basic whitespace cleaning and splitting on a piece of text.""" - text = text.strip() - if not text: - return [] - tokens = text.split() - return tokens - - -class WordpieceTokenizer(object): - """Runs WordPiece tokenization.""" - - def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=100): - self.vocab = vocab - self.unk_token = unk_token - self.max_input_chars_per_word = max_input_chars_per_word - - def tokenize(self, text): - """Tokenizes a piece of text into its word pieces. - - This uses a greedy longest-match-first algorithm to perform tokenization - using the given vocabulary. - - For example: - input = "unaffable" - output = ["un", "##aff", "##able"] - - Args: - text: A single token or whitespace separated tokens. This should have - already been passed through `BasicTokenizer`. - - Returns: - A list of wordpiece tokens. - """ - - output_tokens = [] - for token in whitespace_tokenize(text): - chars = list(token) - if len(chars) > self.max_input_chars_per_word: - output_tokens.append(self.unk_token) - continue - - is_bad = False - start = 0 - sub_tokens = [] - while start < len(chars): - end = len(chars) - cur_substr = None - while start < end: - substr = "".join(chars[start:end]) - if start > 0: - substr = "##" + substr - if substr in self.vocab: - cur_substr = substr - break - end -= 1 - if cur_substr is None: - is_bad = True - break - sub_tokens.append(cur_substr) - start = end - - if is_bad: - output_tokens.append(self.unk_token) - else: - output_tokens.extend(sub_tokens) - return output_tokens - - -def load_vocab(vocab_file): - """Loads a vocabulary file into a dictionary.""" - vocab = collections.OrderedDict() - index = 0 - with open(vocab_file, "r", encoding="utf-8") as reader: - while True: - token = reader.readline() - if not token: - break - token = token.strip() - vocab[token] = index - index += 1 - return vocab - -class BasicTokenizer(object): - """Runs basic tokenization (punctuation splitting, lower casing, etc.).""" - - def __init__(self, - do_lower_case=True, - never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): - """Constructs a BasicTokenizer. - - Args: - do_lower_case: Whether to lower case the input. - """ - self.do_lower_case = do_lower_case - self.never_split = never_split - - def tokenize(self, text): - """Tokenizes a piece of text.""" - text = self._clean_text(text) - # This was added on November 1st, 2018 for the multilingual and Chinese - # models. This is also applied to the English models now, but it doesn't - # matter since the English models were not trained on any Chinese data - # and generally don't have any Chinese data in them (there are Chinese - # characters in the vocabulary because Wikipedia does have some Chinese - # words in the English Wikipedia.). - text = self._tokenize_chinese_chars(text) - orig_tokens = whitespace_tokenize(text) - split_tokens = [] - for token in orig_tokens: - if self.do_lower_case and token not in self.never_split: - token = token.lower() - token = self._run_strip_accents(token) - split_tokens.extend(self._run_split_on_punc(token)) - - output_tokens = whitespace_tokenize(" ".join(split_tokens)) - return output_tokens - - def _run_strip_accents(self, text): - """Strips accents from a piece of text.""" - text = unicodedata.normalize("NFD", text) - output = [] - for char in text: - cat = unicodedata.category(char) - if cat == "Mn": - continue - output.append(char) - return "".join(output) - - def _run_split_on_punc(self, text): - """Splits punctuation on a piece of text.""" - if text in self.never_split: - return [text] - chars = list(text) - i = 0 - start_new_word = True - output = [] - while i < len(chars): - char = chars[i] - if _is_punctuation(char): - output.append([char]) - start_new_word = True - else: - if start_new_word: - output.append([]) - start_new_word = False - output[-1].append(char) - i += 1 - - return ["".join(x) for x in output] - - def _tokenize_chinese_chars(self, text): - """Adds whitespace around any CJK character.""" - output = [] - for char in text: - cp = ord(char) - if self._is_chinese_char(cp): - output.append(" ") - output.append(char) - output.append(" ") - else: - output.append(char) - return "".join(output) - - def _is_chinese_char(self, cp): - """Checks whether CP is the codepoint of a CJK character.""" - # This defines a "chinese character" as anything in the CJK Unicode block: - # https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block) - # - # Note that the CJK Unicode block is NOT all Japanese and Korean characters, - # despite its name. The modern Korean Hangul alphabet is a different block, - # as is Japanese Hiragana and Katakana. Those alphabets are used to write - # space-separated words, so they are not treated specially and handled - # like the all of the other languages. - if ((cp >= 0x4E00 and cp <= 0x9FFF) or # - (cp >= 0x3400 and cp <= 0x4DBF) or # - (cp >= 0x20000 and cp <= 0x2A6DF) or # - (cp >= 0x2A700 and cp <= 0x2B73F) or # - (cp >= 0x2B740 and cp <= 0x2B81F) or # - (cp >= 0x2B820 and cp <= 0x2CEAF) or - (cp >= 0xF900 and cp <= 0xFAFF) or # - (cp >= 0x2F800 and cp <= 0x2FA1F)): # - return True - - return False - - def _clean_text(self, text): - """Performs invalid character removal and whitespace cleanup on text.""" - output = [] - for char in text: - cp = ord(char) - if cp == 0 or cp == 0xfffd or _is_control(char): - continue - if _is_whitespace(char): - output.append(" ") - else: - output.append(char) - return "".join(output) - - -def _is_whitespace(char): - """Checks whether `chars` is a whitespace character.""" - # \t, \n, and \r are technically contorl characters but we treat them - # as whitespace since they are generally considered as such. - if char == " " or char == "\t" or char == "\n" or char == "\r": - return True - cat = unicodedata.category(char) - if cat == "Zs": - return True - return False - - -def _is_control(char): - """Checks whether `chars` is a control character.""" - # These are technically control characters but we count them as whitespace - # characters. - if char == "\t" or char == "\n" or char == "\r": - return False - cat = unicodedata.category(char) - if cat.startswith("C"): - return True - return False - - -def _is_punctuation(char): - """Checks whether `chars` is a punctuation character.""" - cp = ord(char) - # We treat all non-letter/number ASCII as punctuation. - # Characters such as "^", "$", and "`" are not in the Unicode - # Punctuation class but we treat them as punctuation anyways, for - # consistency. - if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or - (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)): - return True - cat = unicodedata.category(char) - if cat.startswith("P"): - return True - return False - - -class BertTokenizer(object): - """Runs end-to-end tokenization: punctuation splitting + wordpiece""" - - def __init__(self, vocab_file, do_lower_case=True, max_len=None, do_basic_tokenize=True, - never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): - """Constructs a BertTokenizer. - - Args: - vocab_file: Path to a one-wordpiece-per-line vocabulary file - do_lower_case: Whether to lower case the input - Only has an effect when do_wordpiece_only=False - do_basic_tokenize: Whether to do basic tokenization before wordpiece. - max_len: An artificial maximum length to truncate tokenized sequences to; - Effective maximum length is always the minimum of this - value (if specified) and the underlying BERT model's - sequence length. - never_split: List of tokens which will never be split during tokenization. - Only has an effect when do_wordpiece_only=False - """ - if not os.path.isfile(vocab_file): - raise ValueError( - "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained " - "model use `tokenizer = BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)) - self.vocab = load_vocab(vocab_file) - self.ids_to_tokens = collections.OrderedDict( - [(ids, tok) for tok, ids in self.vocab.items()]) - self.do_basic_tokenize = do_basic_tokenize - if do_basic_tokenize: - self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case, - never_split=never_split) - self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) - self.max_len = max_len if max_len is not None else int(1e12) - - def _reinit_on_new_vocab(self, vocab): - """ - 在load bert之后，可能会对vocab进行重新排列。重新排列之后调用这个函数重新初始化与vocab相关的性质 - - :param vocab: - :return: - """ - self.vocab = vocab - self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) - - def tokenize(self, text): - split_tokens = [] - if self.do_basic_tokenize: - for token in self.basic_tokenizer.tokenize(text): - for sub_token in self.wordpiece_tokenizer.tokenize(token): - split_tokens.append(sub_token) - else: - split_tokens = self.wordpiece_tokenizer.tokenize(text) - return split_tokens - - def convert_tokens_to_ids(self, tokens): - """Converts a sequence of tokens into ids using the vocab.""" - ids = [] - for token in tokens: - ids.append(self.vocab[token]) - if len(ids) > self.max_len: - print( - "Token indices sequence length is longer than the specified maximum " - " sequence length for this BERT model ({} > {}). Running this" - " sequence through BERT will result in indexing errors".format(len(ids), self.max_len) - ) - return ids - - def convert_ids_to_tokens(self, ids): - """Converts a sequence of ids in wordpiece tokens using the vocab.""" - tokens = [] - for i in ids: - tokens.append(self.ids_to_tokens[i]) - return tokens - - def save_vocabulary(self, vocab_path): - """Save the tokenizer vocabulary to a directory or file.""" - index = 0 - if os.path.isdir(vocab_path): - vocab_file = os.path.join(vocab_path, VOCAB_NAME) - else: - vocab_file = vocab_path - with open(vocab_file, "w", encoding="utf-8") as writer: - for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]): - if index != token_index: - print("Saving vocabulary to {}: vocabulary indices are not consecutive." - " Please check that the vocabulary is not corrupted!".format(vocab_file)) - index = token_index - writer.write(token + u'\n') - index += 1 - return vocab_file - - @classmethod - def from_pretrained(cls, model_dir, *inputs, **kwargs): - """ - 给定path，直接读取vocab. - - """ - pretrained_model_name_or_path = os.path.join(model_dir, VOCAB_NAME) - print("loading vocabulary file {}".format(pretrained_model_name_or_path)) - max_len = 512 - kwargs['max_len'] = min(kwargs.get('max_len', int(1e12)), max_len) - # Instantiate tokenizer. - tokenizer = cls(pretrained_model_name_or_path, *inputs, **kwargs) - return tokenizer - -VOCAB_NAME = 'vocab.txt' - -class _WordBertModel(nn.Module): - def __init__(self, model_dir:str, vocab:Vocabulary, layers:str='-1', pool_method:str='first', include_cls_sep:bool=False): - super().__init__() - - self.tokenzier = BertTokenizer.from_pretrained(model_dir) - self.encoder = BertModel.from_pretrained(model_dir) - # 检查encoder_layer_number是否合理 - encoder_layer_number = len(self.encoder.encoder.layer) - self.layers = list(map(int, layers.split(','))) - for layer in self.layers: - if layer<0: - assert -layer<=encoder_layer_number, f"The layer index:{layer} is out of scope for " \ - f"a bert model with {encoder_layer_number} layers." - else: - assert layer 1 or weight_count > 1: - raise Exception(f"Multiple config files(*.json) or weight files(*.hdf5) detected in {model_dir}.") - elif config_count == 0 or weight_count == 0: - raise Exception(f"No config file or weight file found in {model_dir}") - - config = json.load(open(os.path.join(model_dir, config_file), 'r')) - self.weight_file = os.path.join(model_dir, weight_file) - self.config = config - - OOV_TAG = '' - PAD_TAG = '' - BOS_TAG = '' - EOS_TAG = '' - BOW_TAG = '' - EOW_TAG = '' - - # For the model trained with character-based word encoder. - char_lexicon = {} - with codecs.open(os.path.join(model_dir, 'char.dic'), 'r', encoding='utf-8') as fpi: - for line in fpi: - tokens = line.strip().split('\t') - if len(tokens) == 1: - tokens.insert(0, '\u3000') - token, i = tokens - char_lexicon[token] = int(i) - - # 做一些sanity check - for special_word in [PAD_TAG, OOV_TAG, BOW_TAG, EOW_TAG]: - assert special_word in char_lexicon, f"{special_word} not found in char.dic." - - # 从vocab中构建char_vocab - char_vocab = Vocabulary(unknown=OOV_TAG, padding=PAD_TAG) - # 需要保证与在里面 - char_vocab.add_word_lst([BOW_TAG, EOW_TAG, BOS_TAG, EOS_TAG]) - - for word, index in vocab: - char_vocab.add_word_lst(list(word)) - - self.bos_index, self.eos_index, self._pad_index = len(vocab), len(vocab) + 1, vocab.padding_idx - # 根据char_lexicon调整, 多设置一位，是预留给word padding的(该位置的char表示为全0表示) - char_emb_layer = nn.Embedding(len(char_vocab) + 1, int(config['char_cnn']['embedding']['dim']), - padding_idx=len(char_vocab)) - - # 读入预训练权重 这里的elmo_model 包含char_cnn和 lstm 的 state_dict - elmo_model = torch.load(os.path.join(self.model_dir, weight_file), map_location='cpu') - - char_embed_weights = elmo_model["char_cnn"]['char_emb_layer.weight'] - - found_char_count = 0 - for char, index in char_vocab: # 调整character embedding - if char in char_lexicon: - index_in_pre = char_lexicon.get(char) - found_char_count += 1 - else: - index_in_pre = char_lexicon[OOV_TAG] - char_emb_layer.weight.data[index] = char_embed_weights[index_in_pre] - - print(f"{found_char_count} out of {len(char_vocab)} characters were found in pretrained elmo embedding.") - # 生成words到chars的映射 - max_chars = config['char_cnn']['max_characters_per_token'] - - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab) + 2, max_chars), - fill_value=len(char_vocab), - dtype=torch.long), - requires_grad=False) - for word, index in list(iter(vocab)) + [(BOS_TAG, len(vocab)), (EOS_TAG, len(vocab) + 1)]: - if len(word) + 2 > max_chars: - word = word[:max_chars - 2] - if index == self._pad_index: - continue - elif word == BOS_TAG or word == EOS_TAG: - char_ids = [char_vocab.to_index(BOW_TAG)] + [char_vocab.to_index(word)] + [ - char_vocab.to_index(EOW_TAG)] - char_ids += [char_vocab.to_index(PAD_TAG)] * (max_chars - len(char_ids)) - else: - char_ids = [char_vocab.to_index(BOW_TAG)] + [char_vocab.to_index(c) for c in word] + [ - char_vocab.to_index(EOW_TAG)] - char_ids += [char_vocab.to_index(PAD_TAG)] * (max_chars - len(char_ids)) - self.words_to_chars_embedding[index] = torch.LongTensor(char_ids) - - self.char_vocab = char_vocab - - self.token_embedder = ConvTokenEmbedder( - config, self.weight_file, None, char_emb_layer) - elmo_model["char_cnn"]['char_emb_layer.weight'] = char_emb_layer.weight - self.token_embedder.load_state_dict(elmo_model["char_cnn"]) - - self.output_dim = config['lstm']['projection_dim'] - - # lstm encoder - self.encoder = ElmobiLm(config) - self.encoder.load_state_dict(elmo_model["lstm"]) - - if cache_word_reprs: - if config['char_cnn']['embedding']['dim'] > 0: # 只有在使用了chars的情况下有用 - print("Start to generate cache word representations.") - batch_size = 320 - # bos eos - word_size = self.words_to_chars_embedding.size(0) - num_batches = word_size // batch_size + \ - int(word_size % batch_size != 0) - - self.cached_word_embedding = nn.Embedding(word_size, - config['lstm']['projection_dim']) - with torch.no_grad(): - for i in range(num_batches): - words = torch.arange(i * batch_size, - min((i + 1) * batch_size, word_size)).long() - chars = self.words_to_chars_embedding[words].unsqueeze(1) # batch_size x 1 x max_chars - word_reprs = self.token_embedder(words.unsqueeze(1), - chars).detach() # batch_size x 1 x config['encoder']['projection_dim'] - self.cached_word_embedding.weight.data[words] = word_reprs.squeeze(1) - - print("Finish generating cached word representations. Going to delete the character encoder.") - del self.token_embedder, self.words_to_chars_embedding - else: - print("There is no need to cache word representations, since no character information is used.") - - def forward(self, words): - """ - - :param words: batch_size x max_len - :return: num_layers x batch_size x max_len x hidden_size - """ - # 扩展, - batch_size, max_len = words.size() - expanded_words = words.new_zeros(batch_size, max_len + 2) # 因为pad一定为0， - seq_len = words.ne(self._pad_index).sum(dim=-1) - expanded_words[:, 1:-1] = words - expanded_words[:, 0].fill_(self.bos_index) - expanded_words[torch.arange(batch_size).to(words), seq_len + 1] = self.eos_index - seq_len = seq_len + 2 - zero_tensor = expanded_words.new_zeros(expanded_words.shape) - mask = (expanded_words == zero_tensor).unsqueeze(-1) - if hasattr(self, 'cached_word_embedding'): - token_embedding = self.cached_word_embedding(expanded_words) - else: - if hasattr(self, 'words_to_chars_embedding'): - chars = self.words_to_chars_embedding[expanded_words] - else: - chars = None - token_embedding = self.token_embedder(expanded_words, chars) # batch_size x max_len x embed_dim - - encoder_output = self.encoder(token_embedding, seq_len) - if encoder_output.size(2) < max_len + 2: - num_layers, _, output_len, hidden_size = encoder_output.size() - dummy_tensor = encoder_output.new_zeros(num_layers, batch_size, - max_len + 2 - output_len, hidden_size) - encoder_output = torch.cat((encoder_output, dummy_tensor), 2) - sz = encoder_output.size() # 2, batch_size, max_len, hidden_size - token_embedding = token_embedding.masked_fill(mask, 0) - token_embedding = torch.cat((token_embedding, token_embedding), dim=2).view(1, sz[1], sz[2], sz[3]) - encoder_output = torch.cat((token_embedding, encoder_output), dim=0) - - # 删除, . 这里没有精确地删除，但应该也不会影响最后的结果了。 - encoder_output = encoder_output[:, :, 1:-1] - return encoder_output diff --git a/fastNLP/modules/encoder/attention.py b/fastNLP/modules/encoder/attention.py index 0a42d889..fe3f7fd8 100644 --- a/fastNLP/modules/encoder/attention.py +++ b/fastNLP/modules/encoder/attention.py @@ -8,8 +8,6 @@ import torch import torch.nn.functional as F from torch import nn -from fastNLP.modules.dropout import TimestepDropout - from fastNLP.modules.utils import initial_parameter @@ -18,7 +16,7 @@ class DotAttention(nn.Module): .. todo:: 补上文档 """ - + def __init__(self, key_size, value_size, dropout=0.0): super(DotAttention, self).__init__() self.key_size = key_size @@ -26,7 +24,7 @@ class DotAttention(nn.Module): self.scale = math.sqrt(key_size) self.drop = nn.Dropout(dropout) self.softmax = nn.Softmax(dim=2) - + def forward(self, Q, K, V, mask_out=None): """ @@ -45,7 +43,7 @@ class DotAttention(nn.Module): class MultiHeadAttention(nn.Module): """ - 别名：:class:`fastNLP.modules.MultiHeadAttention` :class:`fastNLP.modules.encoder.attention.MultiHeadAttention` + 别名：:class:`fastNLP.modules.MultiHeadAttention` :class:`fastNLP.modules.encoder.MultiHeadAttention` :param input_size: int, 输入维度的大小。同时也是输出维度的大小。 :param key_size: int, 每个head的维度大小。 @@ -53,14 +51,14 @@ class MultiHeadAttention(nn.Module): :param num_head: int，head的数量。 :param dropout: float。 """ - + def __init__(self, input_size, key_size, value_size, num_head, dropout=0.1): super(MultiHeadAttention, self).__init__() self.input_size = input_size self.key_size = key_size self.value_size = value_size self.num_head = num_head - + in_size = key_size * num_head self.q_in = nn.Linear(input_size, in_size) self.k_in = nn.Linear(input_size, in_size) @@ -69,14 +67,14 @@ class MultiHeadAttention(nn.Module): self.attention = DotAttention(key_size=key_size, value_size=value_size, dropout=dropout) self.out = nn.Linear(value_size * num_head, input_size) self.reset_parameters() - + def reset_parameters(self): sqrt = math.sqrt nn.init.normal_(self.q_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.key_size))) nn.init.normal_(self.k_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.key_size))) nn.init.normal_(self.v_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.value_size))) nn.init.xavier_normal_(self.out.weight) - + def forward(self, Q, K, V, atte_mask_out=None): """ @@ -92,7 +90,7 @@ class MultiHeadAttention(nn.Module): q = self.q_in(Q).view(batch, sq, n_head, d_k) k = self.k_in(K).view(batch, sk, n_head, d_k) v = self.v_in(V).view(batch, sk, n_head, d_v) - + # transpose q, k and v to do batch attention q = q.permute(2, 0, 1, 3).contiguous().view(-1, sq, d_k) k = k.permute(2, 0, 1, 3).contiguous().view(-1, sk, d_k) @@ -100,7 +98,7 @@ class MultiHeadAttention(nn.Module): if atte_mask_out is not None: atte_mask_out = atte_mask_out.repeat(n_head, 1, 1) atte = self.attention(q, k, v, atte_mask_out).view(n_head, batch, sq, d_v) - + # concat all heads, do output linear atte = atte.permute(1, 2, 0, 3).contiguous().view(batch, sq, -1) output = self.out(atte) @@ -124,11 +122,11 @@ class BiAttention(nn.Module): \end{array} """ - + def __init__(self): super(BiAttention, self).__init__() self.inf = 10e12 - + def forward(self, in_x1, in_x2, x1_len, x2_len): """ :param torch.Tensor in_x1: [batch_size, x1_seq_len, hidden_size] 第一句的特征表示 @@ -139,36 +137,36 @@ class BiAttention(nn.Module): torch.Tensor out_x2: [batch_size, x2_seq_len, hidden_size] 第一句attend到的特征表示 """ - + assert in_x1.size()[0] == in_x2.size()[0] assert in_x1.size()[2] == in_x2.size()[2] # The batch size and hidden size must be equal. assert in_x1.size()[1] == x1_len.size()[1] and in_x2.size()[1] == x2_len.size()[1] # The seq len in in_x and x_len must be equal. assert in_x1.size()[0] == x1_len.size()[0] and x1_len.size()[0] == x2_len.size()[0] - + batch_size = in_x1.size()[0] x1_max_len = in_x1.size()[1] x2_max_len = in_x2.size()[1] - + in_x2_t = torch.transpose(in_x2, 1, 2) # [batch_size, hidden_size, x2_seq_len] - + attention_matrix = torch.bmm(in_x1, in_x2_t) # [batch_size, x1_seq_len, x2_seq_len] - + a_mask = x1_len.le(0.5).float() * -self.inf # [batch_size, x1_seq_len] a_mask = a_mask.view(batch_size, x1_max_len, -1) a_mask = a_mask.expand(-1, -1, x2_max_len) # [batch_size, x1_seq_len, x2_seq_len] b_mask = x2_len.le(0.5).float() * -self.inf b_mask = b_mask.view(batch_size, -1, x2_max_len) b_mask = b_mask.expand(-1, x1_max_len, -1) # [batch_size, x1_seq_len, x2_seq_len] - + attention_a = F.softmax(attention_matrix + a_mask, dim=2) # [batch_size, x1_seq_len, x2_seq_len] attention_b = F.softmax(attention_matrix + b_mask, dim=1) # [batch_size, x1_seq_len, x2_seq_len] - + out_x1 = torch.bmm(attention_a, in_x2) # [batch_size, x1_seq_len, hidden_size] attention_b_t = torch.transpose(attention_b, 1, 2) out_x2 = torch.bmm(attention_b_t, in_x1) # [batch_size, x2_seq_len, hidden_size] - + return out_x1, out_x2 @@ -182,10 +180,10 @@ class SelfAttention(nn.Module): :param float drop: dropout概率，默认值为0.5 :param str initial_method: 初始化参数方法 """ - + def __init__(self, input_size, attention_unit=300, attention_hops=10, drop=0.5, initial_method=None, ): super(SelfAttention, self).__init__() - + self.attention_hops = attention_hops self.ws1 = nn.Linear(input_size, attention_unit, bias=False) self.ws2 = nn.Linear(attention_unit, attention_hops, bias=False) @@ -194,7 +192,7 @@ class SelfAttention(nn.Module): self.drop = nn.Dropout(drop) self.tanh = nn.Tanh() initial_parameter(self, initial_method) - + def _penalization(self, attention): """ compute the penalization term for attention module @@ -208,7 +206,7 @@ class SelfAttention(nn.Module): mat = torch.bmm(attention, attention_t) - self.I[:attention.size(0)] ret = (torch.sum(torch.sum((mat ** 2), 2), 1).squeeze() + 1e-10) ** 0.5 return torch.sum(ret) / size[0] - + def forward(self, input, input_origin): """ :param torch.Tensor input: [baz, senLen, h_dim] 要做attention的矩阵 @@ -218,14 +216,14 @@ class SelfAttention(nn.Module): """ input = input.contiguous() size = input.size() # [bsz, len, nhid] - + input_origin = input_origin.expand(self.attention_hops, -1, -1) # [hops,baz, len] input_origin = input_origin.transpose(0, 1).contiguous() # [baz, hops,len] - + y1 = self.tanh(self.ws1(self.drop(input))) # [baz,len,dim] -->[bsz,len, attention-unit] attention = self.ws2(y1).transpose(1, 2).contiguous() # [bsz,len, attention-unit]--> [bsz, len, hop]--> [baz,hop,len] - + attention = attention + (-999999 * (input_origin == 0).float()) # remove the weight on padding token. attention = F.softmax(attention, 2) # [baz ,hop, len] return torch.bmm(attention, input), self._penalization(attention) # output1 --> [baz ,hop ,nhid] diff --git a/fastNLP/modules/encoder/bert.py b/fastNLP/modules/encoder/bert.py index 1819cc69..e73b2c40 100644 --- a/fastNLP/modules/encoder/bert.py +++ b/fastNLP/modules/encoder/bert.py @@ -1,79 +1,925 @@ +""" +这个页面的代码很大程度上参考(复制粘贴)了https://github.com/huggingface/pytorch-pretrained-BERT的代码， 如果你发现该代码对你 + 有用，也请引用一下他们。 +""" +__all__ = [ + "BertModel" +] + +import collections + +import unicodedata +import copy +import json +import math import os -from torch import nn + import torch -from ...io.file_utils import _get_base_url, cached_path, PRETRAINED_BERT_MODEL_DIR -from ._bert import _WordPieceBertModel, BertModel +from torch import nn +import sys +from ..utils import _get_file_name_base_on_postfix -class BertWordPieceEncoder(nn.Module): +CONFIG_FILE = 'bert_config.json' +VOCAB_NAME = 'vocab.txt' + + + +class BertConfig(object): + """Configuration class to store the configuration of a `BertModel`. """ - 读取bert模型，读取之后调用index_dataset方法在dataset中生成word_pieces这一列。 - :param str model_dir_or_name: 模型所在目录或者模型的名称。默认值为``en-base-uncased`` - :param str layers:最终结果中的表示。以','隔开层数，可以以负数去索引倒数几层 - :param bool requires_grad: 是否需要gradient。 + def __init__(self, + vocab_size_or_config_json_file, + hidden_size=768, + num_hidden_layers=12, + num_attention_heads=12, + intermediate_size=3072, + hidden_act="gelu", + hidden_dropout_prob=0.1, + attention_probs_dropout_prob=0.1, + max_position_embeddings=512, + type_vocab_size=2, + initializer_range=0.02, + layer_norm_eps=1e-12): + """Constructs BertConfig. + + Args: + vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in `BertModel`. + hidden_size: Size of the encoder layers and the pooler layer. + num_hidden_layers: Number of hidden layers in the Transformer encoder. + num_attention_heads: Number of attention heads for each attention layer in + the Transformer encoder. + intermediate_size: The size of the "intermediate" (i.e., feed-forward) + layer in the Transformer encoder. + hidden_act: The non-linear activation function (function or string) in the + encoder and pooler. If string, "gelu", "relu" and "swish" are supported. + hidden_dropout_prob: The dropout probabilitiy for all fully connected + layers in the embeddings, encoder, and pooler. + attention_probs_dropout_prob: The dropout ratio for the attention + probabilities. + max_position_embeddings: The maximum sequence length that this model might + ever be used with. Typically set this to something large just in case + (e.g., 512 or 1024 or 2048). + type_vocab_size: The vocabulary size of the `token_type_ids` passed into + `BertModel`. + initializer_range: The sttdev of the truncated_normal_initializer for + initializing all weight matrices. + layer_norm_eps: The epsilon used by LayerNorm. + """ + if isinstance(vocab_size_or_config_json_file, str): + with open(vocab_size_or_config_json_file, "r", encoding='utf-8') as reader: + json_config = json.loads(reader.read()) + for key, value in json_config.items(): + self.__dict__[key] = value + elif isinstance(vocab_size_or_config_json_file, int): + self.vocab_size = vocab_size_or_config_json_file + self.hidden_size = hidden_size + self.num_hidden_layers = num_hidden_layers + self.num_attention_heads = num_attention_heads + self.hidden_act = hidden_act + self.intermediate_size = intermediate_size + self.hidden_dropout_prob = hidden_dropout_prob + self.attention_probs_dropout_prob = attention_probs_dropout_prob + self.max_position_embeddings = max_position_embeddings + self.type_vocab_size = type_vocab_size + self.initializer_range = initializer_range + self.layer_norm_eps = layer_norm_eps + else: + raise ValueError("First argument must be either a vocabulary size (int)" + "or the path to a pretrained model config file (str)") + + @classmethod + def from_dict(cls, json_object): + """Constructs a `BertConfig` from a Python dictionary of parameters.""" + config = BertConfig(vocab_size_or_config_json_file=-1) + for key, value in json_object.items(): + config.__dict__[key] = value + return config + + @classmethod + def from_json_file(cls, json_file): + """Constructs a `BertConfig` from a json file of parameters.""" + with open(json_file, "r", encoding='utf-8') as reader: + text = reader.read() + return cls.from_dict(json.loads(text)) + + def __repr__(self): + return str(self.to_json_string()) + + def to_dict(self): + """Serializes this instance to a Python dictionary.""" + output = copy.deepcopy(self.__dict__) + return output + + def to_json_string(self): + """Serializes this instance to a JSON string.""" + return json.dumps(self.to_dict(), indent=2, sort_keys=True) + "\n" + + def to_json_file(self, json_file_path): + """ Save this instance to a json file.""" + with open(json_file_path, "w", encoding='utf-8') as writer: + writer.write(self.to_json_string()) + + +def gelu(x): + return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0))) + + +def swish(x): + return x * torch.sigmoid(x) + + +ACT2FN = {"gelu": gelu, "relu": torch.nn.functional.relu, "swish": swish} + + +class BertLayerNorm(nn.Module): + def __init__(self, hidden_size, eps=1e-12): + """Construct a layernorm module in the TF style (epsilon inside the square root). + """ + super(BertLayerNorm, self).__init__() + self.weight = nn.Parameter(torch.ones(hidden_size)) + self.bias = nn.Parameter(torch.zeros(hidden_size)) + self.variance_epsilon = eps + + def forward(self, x): + u = x.mean(-1, keepdim=True) + s = (x - u).pow(2).mean(-1, keepdim=True) + x = (x - u) / torch.sqrt(s + self.variance_epsilon) + return self.weight * x + self.bias + + +class BertEmbeddings(nn.Module): + """Construct the embeddings from word, position and token_type embeddings. """ - def __init__(self, model_dir_or_name: str='en-base-uncased', layers: str='-1', - requires_grad: bool=False): - super().__init__() - PRETRAIN_URL = _get_base_url('bert') - - if model_dir_or_name in PRETRAINED_BERT_MODEL_DIR: - model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(model_dir_or_name): - model_dir = model_dir_or_name + + def __init__(self, config): + super(BertEmbeddings, self).__init__() + self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=0) + self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size) + self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size) + + # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load + # any TensorFlow checkpoint file + self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) + self.dropout = nn.Dropout(config.hidden_dropout_prob) + + def forward(self, input_ids, token_type_ids=None): + seq_length = input_ids.size(1) + position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device) + position_ids = position_ids.unsqueeze(0).expand_as(input_ids) + if token_type_ids is None: + token_type_ids = torch.zeros_like(input_ids) + + words_embeddings = self.word_embeddings(input_ids) + position_embeddings = self.position_embeddings(position_ids) + token_type_embeddings = self.token_type_embeddings(token_type_ids) + + embeddings = words_embeddings + position_embeddings + token_type_embeddings + embeddings = self.LayerNorm(embeddings) + embeddings = self.dropout(embeddings) + return embeddings + + +class BertSelfAttention(nn.Module): + def __init__(self, config): + super(BertSelfAttention, self).__init__() + if config.hidden_size % config.num_attention_heads != 0: + raise ValueError( + "The hidden size (%d) is not a multiple of the number of attention " + "heads (%d)" % (config.hidden_size, config.num_attention_heads)) + self.num_attention_heads = config.num_attention_heads + self.attention_head_size = int(config.hidden_size / config.num_attention_heads) + self.all_head_size = self.num_attention_heads * self.attention_head_size + + self.query = nn.Linear(config.hidden_size, self.all_head_size) + self.key = nn.Linear(config.hidden_size, self.all_head_size) + self.value = nn.Linear(config.hidden_size, self.all_head_size) + + self.dropout = nn.Dropout(config.attention_probs_dropout_prob) + + def transpose_for_scores(self, x): + new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size) + x = x.view(*new_x_shape) + return x.permute(0, 2, 1, 3) + + def forward(self, hidden_states, attention_mask): + mixed_query_layer = self.query(hidden_states) + mixed_key_layer = self.key(hidden_states) + mixed_value_layer = self.value(hidden_states) + + query_layer = self.transpose_for_scores(mixed_query_layer) + key_layer = self.transpose_for_scores(mixed_key_layer) + value_layer = self.transpose_for_scores(mixed_value_layer) + + # Take the dot product between "query" and "key" to get the raw attention scores. + attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2)) + attention_scores = attention_scores / math.sqrt(self.attention_head_size) + # Apply the attention mask is (precomputed for all layers in BertModel forward() function) + attention_scores = attention_scores + attention_mask + + # Normalize the attention scores to probabilities. + attention_probs = nn.Softmax(dim=-1)(attention_scores) + + # This is actually dropping out entire tokens to attend to, which might + # seem a bit unusual, but is taken from the original Transformer paper. + attention_probs = self.dropout(attention_probs) + + context_layer = torch.matmul(attention_probs, value_layer) + context_layer = context_layer.permute(0, 2, 1, 3).contiguous() + new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,) + context_layer = context_layer.view(*new_context_layer_shape) + return context_layer + + +class BertSelfOutput(nn.Module): + def __init__(self, config): + super(BertSelfOutput, self).__init__() + self.dense = nn.Linear(config.hidden_size, config.hidden_size) + self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) + self.dropout = nn.Dropout(config.hidden_dropout_prob) + + def forward(self, hidden_states, input_tensor): + hidden_states = self.dense(hidden_states) + hidden_states = self.dropout(hidden_states) + hidden_states = self.LayerNorm(hidden_states + input_tensor) + return hidden_states + + +class BertAttention(nn.Module): + def __init__(self, config): + super(BertAttention, self).__init__() + self.self = BertSelfAttention(config) + self.output = BertSelfOutput(config) + + def forward(self, input_tensor, attention_mask): + self_output = self.self(input_tensor, attention_mask) + attention_output = self.output(self_output, input_tensor) + return attention_output + + +class BertIntermediate(nn.Module): + def __init__(self, config): + super(BertIntermediate, self).__init__() + self.dense = nn.Linear(config.hidden_size, config.intermediate_size) + if isinstance(config.hidden_act, str): + self.intermediate_act_fn = ACT2FN[config.hidden_act] else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") + self.intermediate_act_fn = config.hidden_act + + def forward(self, hidden_states): + hidden_states = self.dense(hidden_states) + hidden_states = self.intermediate_act_fn(hidden_states) + return hidden_states + + +class BertOutput(nn.Module): + def __init__(self, config): + super(BertOutput, self).__init__() + self.dense = nn.Linear(config.intermediate_size, config.hidden_size) + self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps) + self.dropout = nn.Dropout(config.hidden_dropout_prob) + + def forward(self, hidden_states, input_tensor): + hidden_states = self.dense(hidden_states) + hidden_states = self.dropout(hidden_states) + hidden_states = self.LayerNorm(hidden_states + input_tensor) + return hidden_states + + +class BertLayer(nn.Module): + def __init__(self, config): + super(BertLayer, self).__init__() + self.attention = BertAttention(config) + self.intermediate = BertIntermediate(config) + self.output = BertOutput(config) + + def forward(self, hidden_states, attention_mask): + attention_output = self.attention(hidden_states, attention_mask) + intermediate_output = self.intermediate(attention_output) + layer_output = self.output(intermediate_output, attention_output) + return layer_output + + +class BertEncoder(nn.Module): + def __init__(self, config): + super(BertEncoder, self).__init__() + layer = BertLayer(config) + self.layer = nn.ModuleList([copy.deepcopy(layer) for _ in range(config.num_hidden_layers)]) + + def forward(self, hidden_states, attention_mask, output_all_encoded_layers=True): + all_encoder_layers = [] + for layer_module in self.layer: + hidden_states = layer_module(hidden_states, attention_mask) + if output_all_encoded_layers: + all_encoder_layers.append(hidden_states) + if not output_all_encoded_layers: + all_encoder_layers.append(hidden_states) + return all_encoder_layers + + +class BertPooler(nn.Module): + def __init__(self, config): + super(BertPooler, self).__init__() + self.dense = nn.Linear(config.hidden_size, config.hidden_size) + self.activation = nn.Tanh() + + def forward(self, hidden_states): + # We "pool" the model by simply taking the hidden state corresponding + # to the first token. + first_token_tensor = hidden_states[:, 0] + pooled_output = self.dense(first_token_tensor) + pooled_output = self.activation(pooled_output) + return pooled_output + + +class BertModel(nn.Module): + """ + 别名：:class:`fastNLP.modules.BertModel` :class:`fastNLP.modules.encoder.BertModel` + + BERT(Bidirectional Embedding Representations from Transformers). + + 如果你想使用预训练好的权重矩阵，请在以下网址下载. + sources:: + + 'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin", + 'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-pytorch_model.bin", + 'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-pytorch_model.bin", + 'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-pytorch_model.bin", + 'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-pytorch_model.bin", + 'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-pytorch_model.bin", + 'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-pytorch_model.bin", + 'bert-base-german-cased': "https://int-deepset-models-bert.s3.eu-central-1.amazonaws.com/pytorch/bert-base-german-cased-pytorch_model.bin", + 'bert-large-uncased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-pytorch_model.bin", + 'bert-large-cased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-pytorch_model.bin", + 'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-pytorch_model.bin", + 'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-pytorch_model.bin", + 'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-pytorch_model.bin" + + + 用预训练权重矩阵来建立BERT模型:: - self.model = _WordPieceBertModel(model_dir=model_dir, layers=layers) - self._embed_size = len(self.model.layers) * self.model.encoder.hidden_size - self.requires_grad = requires_grad + model = BertModel.from_pretrained("path/to/weights/directory") - @property - def requires_grad(self): + 用随机初始化权重矩阵来建立BERT模型:: + + model = BertModel() + + :param int vocab_size: 词表大小，默认值为30522，为BERT English uncase版本的词表大小 + :param int hidden_size: 隐层大小，默认值为768，为BERT base的版本 + :param int num_hidden_layers: 隐藏层数，默认值为12，为BERT base的版本 + :param int num_attention_heads: 多头注意力头数，默认值为12，为BERT base的版本 + :param int intermediate_size: FFN隐藏层大小，默认值是3072，为BERT base的版本 + :param str hidden_act: FFN隐藏层激活函数，默认值为``gelu`` + :param float hidden_dropout_prob: FFN隐藏层dropout，默认值为0.1 + :param float attention_probs_dropout_prob: Attention层的dropout，默认值为0.1 + :param int max_position_embeddings: 最大的序列长度，默认值为512， + :param int type_vocab_size: 最大segment数量，默认值为2 + :param int initializer_range: 初始化权重范围，默认值为0.02 + """ + + def __init__(self, config, *inputs, **kwargs): + super(BertModel, self).__init__() + if not isinstance(config, BertConfig): + raise ValueError( + "Parameter config in `{}(config)` should be an instance of class `BertConfig`. " + "To create a model from a Google pretrained model use " + "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format( + self.__class__.__name__, self.__class__.__name__ + )) + super(BertModel, self).__init__() + self.config = config + self.hidden_size = self.config.hidden_size + self.embeddings = BertEmbeddings(config) + self.encoder = BertEncoder(config) + self.pooler = BertPooler(config) + self.apply(self.init_bert_weights) + + def init_bert_weights(self, module): + """ Initialize the weights. + """ + if isinstance(module, (nn.Linear, nn.Embedding)): + # Slightly different from the TF version which uses truncated_normal for initialization + # cf https://github.com/pytorch/pytorch/pull/5617 + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + elif isinstance(module, BertLayerNorm): + module.bias.data.zero_() + module.weight.data.fill_(1.0) + if isinstance(module, nn.Linear) and module.bias is not None: + module.bias.data.zero_() + + def forward(self, input_ids, token_type_ids=None, attention_mask=None, output_all_encoded_layers=True): + if attention_mask is None: + attention_mask = torch.ones_like(input_ids) + if token_type_ids is None: + token_type_ids = torch.zeros_like(input_ids) + + # We create a 3D attention mask from a 2D tensor mask. + # Sizes are [batch_size, 1, 1, to_seq_length] + # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length] + # this attention mask is more simple than the triangular masking of causal attention + # used in OpenAI GPT, we just need to prepare the broadcast dimension here. + extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2) + + # Since attention_mask is 1.0 for positions we want to attend and 0.0 for + # masked positions, this operation will create a tensor which is 0.0 for + # positions we want to attend and -10000.0 for masked positions. + # Since we are adding it to the raw scores before the softmax, this is + # effectively the same as removing these entirely. + extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility + extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0 + + embedding_output = self.embeddings(input_ids, token_type_ids) + encoded_layers = self.encoder(embedding_output, + extended_attention_mask, + output_all_encoded_layers=output_all_encoded_layers) + sequence_output = encoded_layers[-1] + pooled_output = self.pooler(sequence_output) + if not output_all_encoded_layers: + encoded_layers = encoded_layers[-1] + return encoded_layers, pooled_output + + @classmethod + def from_pretrained(cls, pretrained_model_dir, *inputs, **kwargs): + state_dict = kwargs.get('state_dict', None) + kwargs.pop('state_dict', None) + kwargs.pop('cache_dir', None) + kwargs.pop('from_tf', None) + # Load config + config_file = _get_file_name_base_on_postfix(pretrained_model_dir, '.json') + config = BertConfig.from_json_file(config_file) + # logger.info("Model config {}".format(config)) + # Instantiate model. + model = cls(config, *inputs, **kwargs) + if state_dict is None: + weights_path = _get_file_name_base_on_postfix(pretrained_model_dir, '.bin') + state_dict = torch.load(weights_path, map_location='cpu') + + old_keys = [] + new_keys = [] + for key in state_dict.keys(): + new_key = None + if 'gamma' in key: + new_key = key.replace('gamma', 'weight') + if 'beta' in key: + new_key = key.replace('beta', 'bias') + if new_key: + old_keys.append(key) + new_keys.append(new_key) + for old_key, new_key in zip(old_keys, new_keys): + state_dict[new_key] = state_dict.pop(old_key) + + missing_keys = [] + unexpected_keys = [] + error_msgs = [] + # copy state_dict so _load_from_state_dict can modify it + metadata = getattr(state_dict, '_metadata', None) + state_dict = state_dict.copy() + if metadata is not None: + state_dict._metadata = metadata + + def load(module, prefix=''): + local_metadata = {} if metadata is None else metadata.get(prefix[:-1], {}) + module._load_from_state_dict( + state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs) + for name, child in module._modules.items(): + if child is not None: + load(child, prefix + name + '.') + + load(model, prefix='' if hasattr(model, 'bert') else 'bert.') + if len(missing_keys) > 0: + print("Weights of {} not initialized from pretrained model: {}".format( + model.__class__.__name__, missing_keys)) + if len(unexpected_keys) > 0: + print("Weights from pretrained model not used in {}: {}".format( + model.__class__.__name__, unexpected_keys)) + return model + + +def whitespace_tokenize(text): + """Runs basic whitespace cleaning and splitting on a piece of text.""" + text = text.strip() + if not text: + return [] + tokens = text.split() + return tokens + + +class WordpieceTokenizer(object): + """Runs WordPiece tokenization.""" + + def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=100): + self.vocab = vocab + self.unk_token = unk_token + self.max_input_chars_per_word = max_input_chars_per_word + + def tokenize(self, text): + """Tokenizes a piece of text into its word pieces. + + This uses a greedy longest-match-first algorithm to perform tokenization + using the given vocabulary. + + For example: + input = "unaffable" + output = ["un", "##aff", "##able"] + + Args: + text: A single token or whitespace separated tokens. This should have + already been passed through `BasicTokenizer`. + + Returns: + A list of wordpiece tokens. """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 + + output_tokens = [] + for token in whitespace_tokenize(text): + chars = list(token) + if len(chars) > self.max_input_chars_per_word: + output_tokens.append(self.unk_token) + continue + + is_bad = False + start = 0 + sub_tokens = [] + while start < len(chars): + end = len(chars) + cur_substr = None + while start < end: + substr = "".join(chars[start:end]) + if start > 0: + substr = "##" + substr + if substr in self.vocab: + cur_substr = substr + break + end -= 1 + if cur_substr is None: + is_bad = True + break + sub_tokens.append(cur_substr) + start = end + + if is_bad: + output_tokens.append(self.unk_token) + else: + output_tokens.extend(sub_tokens) + if len(output_tokens)==0: #防止里面全是空格或者回车符号 + return [self.unk_token] + return output_tokens + + +def load_vocab(vocab_file): + """Loads a vocabulary file into a dictionary.""" + vocab = collections.OrderedDict() + index = 0 + with open(vocab_file, "r", encoding="utf-8") as reader: + while True: + token = reader.readline() + if not token: + break + token = token.strip() + vocab[token] = index + index += 1 + return vocab + + +class BasicTokenizer(object): + """Runs basic tokenization (punctuation splitting, lower casing, etc.).""" + + def __init__(self, + do_lower_case=True, + never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): + """Constructs a BasicTokenizer. + + Args: + do_lower_case: Whether to lower case the input. + """ + self.do_lower_case = do_lower_case + self.never_split = never_split + + def tokenize(self, text): + """Tokenizes a piece of text.""" + text = self._clean_text(text) + # This was added on November 1st, 2018 for the multilingual and Chinese + # models. This is also applied to the English models now, but it doesn't + # matter since the English models were not trained on any Chinese data + # and generally don't have any Chinese data in them (there are Chinese + # characters in the vocabulary because Wikipedia does have some Chinese + # words in the English Wikipedia.). + text = self._tokenize_chinese_chars(text) + orig_tokens = whitespace_tokenize(text) + split_tokens = [] + for token in orig_tokens: + if self.do_lower_case and token not in self.never_split: + token = token.lower() + token = self._run_strip_accents(token) + split_tokens.extend(self._run_split_on_punc(token)) + + output_tokens = whitespace_tokenize(" ".join(split_tokens)) + return output_tokens + + def _run_strip_accents(self, text): + """Strips accents from a piece of text.""" + text = unicodedata.normalize("NFD", text) + output = [] + for char in text: + cat = unicodedata.category(char) + if cat == "Mn": + continue + output.append(char) + return "".join(output) + + def _run_split_on_punc(self, text): + """Splits punctuation on a piece of text.""" + if text in self.never_split: + return [text] + chars = list(text) + i = 0 + start_new_word = True + output = [] + while i < len(chars): + char = chars[i] + if _is_punctuation(char): + output.append([char]) + start_new_word = True + else: + if start_new_word: + output.append([]) + start_new_word = False + output[-1].append(char) + i += 1 + + return ["".join(x) for x in output] + + def _tokenize_chinese_chars(self, text): + """Adds whitespace around any CJK character.""" + output = [] + for char in text: + cp = ord(char) + if self._is_chinese_char(cp): + output.append(" ") + output.append(char) + output.append(" ") + else: + output.append(char) + return "".join(output) + + def _is_chinese_char(self, cp): + """Checks whether CP is the codepoint of a CJK character.""" + # This defines a "chinese character" as anything in the CJK Unicode block: + # https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block) + # + # Note that the CJK Unicode block is NOT all Japanese and Korean characters, + # despite its name. The modern Korean Hangul alphabet is a different block, + # as is Japanese Hiragana and Katakana. Those alphabets are used to write + # space-separated words, so they are not treated specially and handled + # like the all of the other languages. + if ((cp >= 0x4E00 and cp <= 0x9FFF) or # + (cp >= 0x3400 and cp <= 0x4DBF) or # + (cp >= 0x20000 and cp <= 0x2A6DF) or # + (cp >= 0x2A700 and cp <= 0x2B73F) or # + (cp >= 0x2B740 and cp <= 0x2B81F) or # + (cp >= 0x2B820 and cp <= 0x2CEAF) or + (cp >= 0xF900 and cp <= 0xFAFF) or # + (cp >= 0x2F800 and cp <= 0x2FA1F)): # + return True + + return False + + def _clean_text(self, text): + """Performs invalid character removal and whitespace cleanup on text.""" + output = [] + for char in text: + cp = ord(char) + if cp == 0 or cp == 0xfffd or _is_control(char): + continue + if _is_whitespace(char): + output.append(" ") + else: + output.append(char) + return "".join(output) + + +def _is_whitespace(char): + """Checks whether `chars` is a whitespace character.""" + # \t, \n, and \r are technically contorl characters but we treat them + # as whitespace since they are generally considered as such. + if char == " " or char == "\t" or char == "\n" or char == "\r": + return True + cat = unicodedata.category(char) + if cat == "Zs": + return True + return False + + +def _is_control(char): + """Checks whether `chars` is a control character.""" + # These are technically control characters but we count them as whitespace + # characters. + if char == "\t" or char == "\n" or char == "\r": + return False + cat = unicodedata.category(char) + if cat.startswith("C"): + return True + return False + + +def _is_punctuation(char): + """Checks whether `chars` is a punctuation character.""" + cp = ord(char) + # We treat all non-letter/number ASCII as punctuation. + # Characters such as "^", "$", and "`" are not in the Unicode + # Punctuation class but we treat them as punctuation anyways, for + # consistency. + if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or + (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)): + return True + cat = unicodedata.category(char) + if cat.startswith("P"): + return True + return False + + +class BertTokenizer(object): + """Runs end-to-end tokenization: punctuation splitting + wordpiece""" + + def __init__(self, vocab_file, do_lower_case=True, max_len=None, do_basic_tokenize=True, + never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): + """Constructs a BertTokenizer. + + Args: + vocab_file: Path to a one-wordpiece-per-line vocabulary file + do_lower_case: Whether to lower case the input + Only has an effect when do_wordpiece_only=False + do_basic_tokenize: Whether to do basic tokenization before wordpiece. + max_len: An artificial maximum length to truncate tokenized sequences to; + Effective maximum length is always the minimum of this + value (if specified) and the underlying BERT model's + sequence length. + never_split: List of tokens which will never be split during tokenization. + Only has an effect when do_wordpiece_only=False + """ + if not os.path.isfile(vocab_file): + raise ValueError( + "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained " + "model use `tokenizer = BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)) + self.vocab = load_vocab(vocab_file) + self.ids_to_tokens = collections.OrderedDict( + [(ids, tok) for tok, ids in self.vocab.items()]) + self.do_basic_tokenize = do_basic_tokenize + if do_basic_tokenize: + self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case, + never_split=never_split) + self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) + self.max_len = max_len if max_len is not None else int(1e12) + + def _reinit_on_new_vocab(self, vocab): + """ + 在load bert之后，可能会对vocab进行重新排列。重新排列之后调用这个函数重新初始化与vocab相关的性质 + + :param vocab: :return: """ - requires_grads = set([param.requires_grad for name, param in self.named_parameters()]) - if len(requires_grads)==1: - return requires_grads.pop() + self.vocab = vocab + self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) + + def tokenize(self, text): + split_tokens = [] + if self.do_basic_tokenize: + for token in self.basic_tokenizer.tokenize(text): + for sub_token in self.wordpiece_tokenizer.tokenize(token): + split_tokens.append(sub_token) else: - return None + split_tokens = self.wordpiece_tokenizer.tokenize(text) + return split_tokens + + def convert_tokens_to_ids(self, tokens): + """Converts a sequence of tokens into ids using the vocab.""" + ids = [] + for token in tokens: + ids.append(self.vocab[token]) + if len(ids) > self.max_len: + print( + "Token indices sequence length is longer than the specified maximum " + " sequence length for this BERT model ({} > {}). Running this" + " sequence through BERT will result in indexing errors".format(len(ids), self.max_len) + ) + return ids + + def convert_ids_to_tokens(self, ids): + """Converts a sequence of ids in wordpiece tokens using the vocab.""" + tokens = [] + for i in ids: + tokens.append(self.ids_to_tokens[i]) + return tokens + + def save_vocabulary(self, vocab_path): + """Save the tokenizer vocabulary to a directory or file.""" + index = 0 + if os.path.isdir(vocab_path): + vocab_file = os.path.join(vocab_path, VOCAB_NAME) + else: + vocab_file = vocab_path + with open(vocab_file, "w", encoding="utf-8") as writer: + for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]): + if index != token_index: + print("Saving vocabulary to {}: vocabulary indices are not consecutive." + " Please check that the vocabulary is not corrupted!".format(vocab_file)) + index = token_index + writer.write(token + u'\n') + index += 1 + return vocab_file + + @classmethod + def from_pretrained(cls, model_dir, *inputs, **kwargs): + """ + 给定path，直接读取vocab. + + """ + pretrained_model_name_or_path = _get_file_name_base_on_postfix(model_dir, '.txt') + print("loading vocabulary file {}".format(pretrained_model_name_or_path)) + max_len = 512 + kwargs['max_len'] = min(kwargs.get('max_position_embeddings', int(1e12)), max_len) + # Instantiate tokenizer. + tokenizer = cls(pretrained_model_name_or_path, *inputs, **kwargs) + return tokenizer - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - param.requires_grad = value +class _WordPieceBertModel(nn.Module): + """ + 这个模块用于直接计算word_piece的结果. - @property - def embed_size(self): - return self._embed_size + """ + + def __init__(self, model_dir: str, layers: str = '-1', pooled_cls:bool=False): + super().__init__() - def index_datasets(self, *datasets, field_name): + self.tokenzier = BertTokenizer.from_pretrained(model_dir) + self.encoder = BertModel.from_pretrained(model_dir) + # 检查encoder_layer_number是否合理 + encoder_layer_number = len(self.encoder.encoder.layer) + self.layers = list(map(int, layers.split(','))) + for layer in self.layers: + if layer < 0: + assert -layer <= encoder_layer_number, f"The layer index:{layer} is out of scope for " \ + f"a bert model with {encoder_layer_number} layers." + else: + assert layer < encoder_layer_number, f"The layer index:{layer} is out of scope for " \ + f"a bert model with {encoder_layer_number} layers." + + self._cls_index = self.tokenzier.vocab['[CLS]'] + self._sep_index = self.tokenzier.vocab['[SEP]'] + self._wordpiece_pad_index = self.tokenzier.vocab['[PAD]'] # 需要用于生成word_piece + self.pooled_cls = pooled_cls + + def index_dataset(self, *datasets, field_name, add_cls_sep=True): """ 使用bert的tokenizer新生成word_pieces列加入到datasets中，并将他们设置为input。如果首尾不是 [CLS]与[SEP]会在首尾额外加入[CLS]与[SEP], 且将word_pieces这一列的pad value设置为了bert的pad value。 :param datasets: DataSet对象 - :param field_name: 基于哪一列的内容生成word_pieces列。这一列中每个数据应该是List[str]的形式。 + :param field_name: 基于哪一列index :return: """ - self.model.index_dataset(*datasets, field_name=field_name) + + def convert_words_to_word_pieces(words): + word_pieces = [] + for word in words: + tokens = self.tokenzier.wordpiece_tokenizer.tokenize(word) + word_piece_ids = self.tokenzier.convert_tokens_to_ids(tokens) + word_pieces.extend(word_piece_ids) + if add_cls_sep: + if word_pieces[0] != self._cls_index: + word_pieces.insert(0, self._cls_index) + if word_pieces[-1] != self._sep_index: + word_pieces.insert(-1, self._sep_index) + return word_pieces + + for index, dataset in enumerate(datasets): + try: + dataset.apply_field(convert_words_to_word_pieces, field_name=field_name, new_field_name='word_pieces', + is_input=True) + dataset.set_pad_val('word_pieces', self._wordpiece_pad_index) + except Exception as e: + print(f"Exception happens when processing the {index} dataset.") + raise e def forward(self, word_pieces, token_type_ids=None): """ - 计算words的bert embedding表示。传入的words中应该自行包含[CLS]与[SEP]的tag。 - :param words: batch_size x max_len - :param token_type_ids: batch_size x max_len, 用于区分前一句和后一句话 - :return: torch.FloatTensor. batch_size x max_len x (768*len(self.layers)) + :param word_pieces: torch.LongTensor, batch_size x max_len + :param token_type_ids: torch.LongTensor, batch_size x max_len + :return: num_layers x batch_size x max_len x hidden_size或者num_layers x batch_size x (max_len+2) x hidden_size """ - outputs = self.model(word_pieces, token_type_ids) - outputs = torch.cat([*outputs], dim=-1) + batch_size, max_len = word_pieces.size() + attn_masks = word_pieces.ne(self._wordpiece_pad_index) + bert_outputs, pooled_cls = self.encoder(word_pieces, token_type_ids=token_type_ids, attention_mask=attn_masks, + output_all_encoded_layers=True) + # output_layers = [self.layers] # len(self.layers) x batch_size x max_word_piece_length x hidden_size + outputs = bert_outputs[0].new_zeros((len(self.layers), batch_size, max_len, bert_outputs[0].size(-1))) + for l_index, l in enumerate(self.layers): + bert_output = bert_outputs[l] + if l==len(bert_outputs) and self.pooled_cls: + bert_output[:, 0] = pooled_cls + outputs[l_index] = bert_output return outputs diff --git a/fastNLP/modules/encoder/char_encoder.py b/fastNLP/modules/encoder/char_encoder.py index 6ce63d1a..6a6e1470 100644 --- a/fastNLP/modules/encoder/char_encoder.py +++ b/fastNLP/modules/encoder/char_encoder.py @@ -11,7 +11,7 @@ from ..utils import initial_parameter # from torch.nn.init import xavier_uniform class ConvolutionCharEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.ConvolutionCharEncoder` :class:`fastNLP.modules.encoder.char_encoder.ConvolutionCharEncoder` + 别名：:class:`fastNLP.modules.ConvolutionCharEncoder` :class:`fastNLP.modules.encoder.ConvolutionCharEncoder` char级别的卷积编码器. @@ -21,15 +21,16 @@ class ConvolutionCharEncoder(nn.Module): :param tuple kernels: 一个由int组成的tuple. tuple的长度是char级别卷积操作的数目, 第`i`个int表示第`i`个卷积操作的卷积核. :param initial_method: 初始化参数的方式, 默认为`xavier normal` """ - + def __init__(self, char_emb_size=50, feature_maps=(40, 30, 30), kernels=(1, 3, 5), initial_method=None): super(ConvolutionCharEncoder, self).__init__() self.convs = nn.ModuleList([ - nn.Conv2d(1, feature_maps[i], kernel_size=(char_emb_size, kernels[i]), bias=True, padding=(0, kernels[i]//2)) + nn.Conv2d(1, feature_maps[i], kernel_size=(char_emb_size, kernels[i]), bias=True, + padding=(0, kernels[i] // 2)) for i in range(len(kernels))]) - + initial_parameter(self, initial_method) - + def forward(self, x): """ :param torch.Tensor x: ``[batch_size * sent_length, word_length, char_emb_size]`` 输入字符的embedding @@ -40,7 +41,7 @@ class ConvolutionCharEncoder(nn.Module): x = x.transpose(2, 3) # [batch_size*sent_length, channel, height, width] return self._convolute(x).unsqueeze(2) - + def _convolute(self, x): feats = [] for conv in self.convs: @@ -57,13 +58,13 @@ class ConvolutionCharEncoder(nn.Module): class LSTMCharEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.LSTMCharEncoder` :class:`fastNLP.modules.encoder.char_encoder.LSTMCharEncoder` + 别名：:class:`fastNLP.modules.LSTMCharEncoder` :class:`fastNLP.modules.encoder.LSTMCharEncoder` char级别基于LSTM的encoder. """ - + def __init__(self, char_emb_size=50, hidden_size=None, initial_method=None): """ :param int char_emb_size: char级别embedding的维度. Default: 50 @@ -73,14 +74,14 @@ class LSTMCharEncoder(nn.Module): """ super(LSTMCharEncoder, self).__init__() self.hidden_size = char_emb_size if hidden_size is None else hidden_size - + self.lstm = nn.LSTM(input_size=char_emb_size, hidden_size=self.hidden_size, num_layers=1, bias=True, batch_first=True) initial_parameter(self, initial_method) - + def forward(self, x): """ :param torch.Tensor x: ``[ n_batch*n_word, word_length, char_emb_size]`` 输入字符的embedding @@ -91,6 +92,6 @@ class LSTMCharEncoder(nn.Module): h0 = nn.init.orthogonal_(h0) c0 = torch.empty(1, batch_size, self.hidden_size) c0 = nn.init.orthogonal_(c0) - + _, hidden = self.lstm(x, (h0, c0)) return hidden[0].squeeze().unsqueeze(2) diff --git a/fastNLP/modules/encoder/conv_maxpool.py b/fastNLP/modules/encoder/conv_maxpool.py index 68605c98..8ce6b163 100644 --- a/fastNLP/modules/encoder/conv_maxpool.py +++ b/fastNLP/modules/encoder/conv_maxpool.py @@ -5,9 +5,10 @@ import torch import torch.nn as nn import torch.nn.functional as F + class ConvMaxpool(nn.Module): """ - 别名：:class:`fastNLP.modules.ConvMaxpool` :class:`fastNLP.modules.encoder.conv_maxpool.ConvMaxpool` + 别名：:class:`fastNLP.modules.ConvMaxpool` :class:`fastNLP.modules.encoder.ConvMaxpool` 集合了Convolution和Max-Pooling于一体的层。给定一个batch_size x max_len x input_size的输入，返回batch_size x sum(output_channels) 大小的matrix。在内部，是先使用CNN给输入做卷积，然后经过activation激活层，在通过在长度(max_len) @@ -18,12 +19,12 @@ class ConvMaxpool(nn.Module): :param int,tuple(int) kernel_sizes: 输出channel的kernel大小。 :param str activation: Convolution后的结果将通过该activation后再经过max-pooling。支持relu, sigmoid, tanh """ - + def __init__(self, in_channels, out_channels, kernel_sizes, activation="relu"): super(ConvMaxpool, self).__init__() for kernel_size in kernel_sizes: - assert kernel_size%2==1, "kernel size has to be odd numbers." + assert kernel_size % 2 == 1, "kernel size has to be odd numbers." # convolution if isinstance(kernel_sizes, (list, tuple, int)): @@ -36,22 +37,22 @@ class ConvMaxpool(nn.Module): " of kernel_sizes." else: raise ValueError("The type of out_channels and kernel_sizes should be the same.") - + self.convs = nn.ModuleList([nn.Conv1d( in_channels=in_channels, out_channels=oc, kernel_size=ks, stride=1, - padding=ks//2, + padding=ks // 2, dilation=1, groups=1, bias=None) for oc, ks in zip(out_channels, kernel_sizes)]) - + else: raise Exception( 'Incorrect kernel sizes: should be list, tuple or int') - + # activation function if activation == 'relu': self.activation = F.relu diff --git a/fastNLP/modules/encoder/embedding.py b/fastNLP/modules/encoder/embedding.py deleted file mode 100644 index 050a423a..00000000 --- a/fastNLP/modules/encoder/embedding.py +++ /dev/null @@ -1,1083 +0,0 @@ -__all__ = [ - "Embedding", - "StaticEmbedding", - "ElmoEmbedding", - "BertEmbedding", - "StackEmbedding", - "LSTMCharEmbedding", - "CNNCharEmbedding", -] -import torch.nn as nn -from ..utils import get_embeddings -from .lstm import LSTM -from ...core.vocabulary import Vocabulary -from abc import abstractmethod -import torch -import numpy as np -import torch.nn.functional as F -import os -from ._elmo import _ElmoModel -from ...io.file_utils import cached_path, _get_base_url -from ._bert import _WordBertModel -from typing import List - -import warnings -from ...core.dataset import DataSet -from ...core.batch import DataSetIter -from ...core.sampler import SequentialSampler -from ...core.utils import _move_model_to_device, _get_model_device -from ...io.file_utils import PRETRAINED_BERT_MODEL_DIR, PRETRAINED_ELMO_MODEL_DIR, PRETRAIN_STATIC_FILES - - -class Embedding(nn.Module): - """ - 别名：:class:`fastNLP.modules.Embedding` :class:`fastNLP.modules.encoder.embedding.Embedding` - - Embedding组件. 可以通过self.num_embeddings获取词表大小; self.embedding_dim获取embedding的维度""" - - def __init__(self, init_embed, word_dropout=0, dropout=0.0, unk_index=None): - """ - - :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray init_embed: Embedding的大小(传入tuple(int, int), - 第一个int为vocab_zie, 第二个int为embed_dim); 如果为Tensor, Embedding, ndarray等则直接使用该值初始化Embedding; - :param float word_dropout: 按照一定概率随机将word设置为unk_index，这样可以使得unk这个token得到足够的训练, 且会对网络有 - 一定的regularize的作用。 - :param float dropout: 对Embedding的输出的dropout。 - :param int unk_index: drop word时替换为的index。fastNLP的Vocabulary的unk_index默认为1。 - """ - super(Embedding, self).__init__() - - self.embed = get_embeddings(init_embed) - - self.dropout = nn.Dropout(dropout) - if not isinstance(self.embed, TokenEmbedding): - self._embed_size = self.embed.weight.size(1) - if word_dropout>0 and not isinstance(unk_index, int): - raise ValueError("When drop word is set, you need to pass in the unk_index.") - else: - self._embed_size = self.embed.embed_size - unk_index = self.embed.get_word_vocab().unknown_idx - self.unk_index = unk_index - self.word_dropout = word_dropout - - def forward(self, x): - """ - :param torch.LongTensor x: [batch, seq_len] - :return: torch.Tensor : [batch, seq_len, embed_dim] - """ - if self.word_dropout>0 and self.training: - mask = torch.ones_like(x).float() * self.word_dropout - mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 - x = x.masked_fill(mask, self.unk_index) - x = self.embed(x) - return self.dropout(x) - - @property - def num_embedding(self)->int: - if isinstance(self.embed, nn.Embedding): - return self.embed.weight.size(0) - else: - return self.embed.num_embedding - - def __len__(self): - return len(self.embed) - - @property - def embed_size(self) -> int: - return self._embed_size - - @property - def embedding_dim(self) -> int: - return self._embed_size - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - if not isinstance(self.embed, TokenEmbedding): - return self.embed.weight.requires_grad - else: - return self.embed.requires_grad - - @requires_grad.setter - def requires_grad(self, value): - if not isinstance(self.embed, TokenEmbedding): - self.embed.weight.requires_grad = value - else: - self.embed.requires_grad = value - - @property - def size(self): - if isinstance(self.embed, TokenEmbedding): - return self.embed.size - else: - return self.embed.weight.size() - - -class TokenEmbedding(nn.Module): - def __init__(self, vocab, word_dropout=0.0, dropout=0.0): - super(TokenEmbedding, self).__init__() - assert vocab.padding is not None, "Vocabulary must have a padding entry." - self._word_vocab = vocab - self._word_pad_index = vocab.padding_idx - if word_dropout>0: - assert vocab.unknown is not None, "Vocabulary must have unknown entry when you want to drop a word." - self.word_dropout = word_dropout - self._word_unk_index = vocab.unknown_idx - self.dropout_layer = nn.Dropout(dropout) - - def drop_word(self, words): - """ - 按照设定随机将words设置为unknown_index。 - - :param torch.LongTensor words: batch_size x max_len - :return: - """ - if self.word_dropout > 0 and self.training: - mask = torch.ones_like(words).float() * self.word_dropout - mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 - words = words.masked_fill(mask, self._word_unk_index) - return words - - def dropout(self, words): - """ - 对embedding后的word表示进行drop。 - - :param torch.FloatTensor words: batch_size x max_len x embed_size - :return: - """ - return self.dropout_layer(words) - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - requires_grads = set([param.requires_grad for param in self.parameters()]) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for param in self.parameters(): - param.requires_grad = value - - def __len__(self): - return len(self._word_vocab) - - @property - def embed_size(self) -> int: - return self._embed_size - - @property - def embedding_dim(self) -> int: - return self._embed_size - - @property - def num_embedding(self) -> int: - """ - 这个值可能会大于实际的embedding矩阵的大小。 - :return: - """ - return len(self._word_vocab) - - def get_word_vocab(self): - """ - 返回embedding的词典。 - - :return: Vocabulary - """ - return self._word_vocab - - @property - def size(self): - return torch.Size(self.num_embedding, self._embed_size) - - @abstractmethod - def forward(self, *input): - raise NotImplementedError - -class StaticEmbedding(TokenEmbedding): - """ - 别名：:class:`fastNLP.modules.StaticEmbedding` :class:`fastNLP.modules.encoder.embedding.StaticEmbedding` - - StaticEmbedding组件. 给定embedding的名称，根据vocab从embedding中抽取相应的数据。该Embedding可以就按照正常的embedding使用了 - - Example:: - - >>> embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50') - - - :param vocab: Vocabulary. 若该项为None则会读取所有的embedding。 - :param model_dir_or_name: 可以有两种方式调用预训练好的static embedding：第一种是传入embedding的文件名，第二种是传入embedding - 的名称。目前支持的embedding包括{`en` 或者 `en-glove-840b-300` : glove.840B.300d, `en-glove-6b-50` : glove.6B.50d, - `en-word2vec-300` : GoogleNews-vectors-negative300}。第二种情况将自动查看缓存中是否存在该模型，没有的话将自动下载。 - :param bool requires_grad: 是否需要gradient. 默认为True - :param callable init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。调用该方法时传入一个tensor对象。 - :param bool lower: 是否将vocab中的词语小写后再和预训练的词表进行匹配。如果你的词表中包含大写的词语，或者就是需要单独 - 为大写的词语开辟一个vector表示，则将lower设置为False。 - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 - :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 - :param bool normailize: 是否对vector进行normalize，使得每个vector的norm为1。 - """ - def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en', requires_grad: bool=True, init_method=None, - lower=False, dropout=0, word_dropout=0, normalize=False): - super(StaticEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - # 得到cache_path - if model_dir_or_name.lower() in PRETRAIN_STATIC_FILES: - PRETRAIN_URL = _get_base_url('static') - model_name = PRETRAIN_STATIC_FILES[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_path = cached_path(model_url) - # 检查是否存在 - elif os.path.isfile(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_path = model_dir_or_name - else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") - - # 读取embedding - if lower: - lowered_vocab = Vocabulary(padding=vocab.padding, unknown=vocab.unknown) - for word, index in vocab: - if not vocab._is_word_no_create_entry(word): - lowered_vocab.add_word(word.lower()) # 先加入需要创建entry的 - for word in vocab._no_create_word.keys(): # 不需要创建entry的 - if word in vocab: - lowered_word = word.lower() - if lowered_word not in lowered_vocab.word_count: - lowered_vocab.add_word(lowered_word) - lowered_vocab._no_create_word[lowered_word] += 1 - print(f"All word in vocab have been lowered. There are {len(vocab)} words, {len(lowered_vocab)} unique lowered " - f"words.") - embedding = self._load_with_vocab(model_path, vocab=lowered_vocab, init_method=init_method, - normalize=normalize) - # 需要适配一下 - if not hasattr(self, 'words_to_words'): - self.words_to_words = torch.arange(len(lowered_vocab, )).long() - if lowered_vocab.unknown: - unknown_idx = lowered_vocab.unknown_idx - else: - unknown_idx = embedding.size(0) - 1 # 否则是最后一个为unknow - words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), - requires_grad=False) - for word, index in vocab: - if word not in lowered_vocab: - word = word.lower() - if lowered_vocab._is_word_no_create_entry(word): # 如果不需要创建entry,已经默认unknown了 - continue - words_to_words[index] = self.words_to_words[lowered_vocab.to_index(word)] - self.words_to_words = words_to_words - else: - embedding = self._load_with_vocab(model_path, vocab=vocab, init_method=init_method, - normalize=normalize) - self.embedding = nn.Embedding(num_embeddings=embedding.shape[0], embedding_dim=embedding.shape[1], - padding_idx=vocab.padding_idx, - max_norm=None, norm_type=2, scale_grad_by_freq=False, - sparse=False, _weight=embedding) - self._embed_size = self.embedding.weight.size(1) - self.requires_grad = requires_grad - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - requires_grads = set([param.requires_grad for name, param in self.named_parameters() - if 'words_to_words' not in name]) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - if 'words_to_words' in name: - continue - param.requires_grad = value - - def _load_with_vocab(self, embed_filepath, vocab, dtype=np.float32, padding='', unknown='', - normalize=True, error='ignore', init_method=None): - """ - 从embed_filepath这个预训练的词向量中抽取出vocab这个词表的词的embedding。EmbedLoader将自动判断embed_filepath是 - word2vec(第一行只有两个元素)还是glove格式的数据。 - - :param str embed_filepath: 预训练的embedding的路径。 - :param vocab: 词表 :class:`~fastNLP.Vocabulary` 类型，读取出现在vocab中的词的embedding。 - 没有出现在vocab中的词的embedding将通过找到的词的embedding的正态分布采样出来，以使得整个Embedding是同分布的。 - :param dtype: 读出的embedding的类型 - :param str padding: 词表中padding的token - :param str unknown: 词表中unknown的token - :param bool normalize: 是否将每个vector归一化到norm为1 - :param str error: `ignore` , `strict` ; 如果 `ignore` ，错误将自动跳过; 如果 `strict` , 错误将抛出。 - 这里主要可能出错的地方在于词表有空行或者词表出现了维度不一致。 - :param init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。默认使用torch.nn.init.zeros_ - :return torch.tensor: shape为 [len(vocab), dimension], dimension由pretrain的embedding决定。 - """ - assert isinstance(vocab, Vocabulary), "Only fastNLP.Vocabulary is supported." - if not os.path.exists(embed_filepath): - raise FileNotFoundError("`{}` does not exist.".format(embed_filepath)) - with open(embed_filepath, 'r', encoding='utf-8') as f: - line = f.readline().strip() - parts = line.split() - start_idx = 0 - if len(parts) == 2: - dim = int(parts[1]) - start_idx += 1 - else: - dim = len(parts) - 1 - f.seek(0) - matrix = {} - found_count = 0 - for idx, line in enumerate(f, start_idx): - try: - parts = line.strip().split() - word = ''.join(parts[:-dim]) - nums = parts[-dim:] - # 对齐unk与pad - if word == padding and vocab.padding is not None: - word = vocab.padding - elif word == unknown and vocab.unknown is not None: - word = vocab.unknown - if word in vocab: - index = vocab.to_index(word) - matrix[index] = torch.from_numpy(np.fromstring(' '.join(nums), sep=' ', dtype=dtype, count=dim)) - found_count += 1 - except Exception as e: - if error == 'ignore': - warnings.warn("Error occurred at the {} line.".format(idx)) - else: - print("Error occurred at the {} line.".format(idx)) - raise e - print("Found {} out of {} words in the pre-training embedding.".format(found_count, len(vocab))) - for word, index in vocab: - if index not in matrix and not vocab._is_word_no_create_entry(word): - if vocab.unknown_idx in matrix: # 如果有unkonwn，用unknown初始化 - matrix[index] = matrix[vocab.unknown_idx] - else: - matrix[index] = None - - vectors = torch.zeros(len(matrix), dim) - if init_method: - init_method(vectors) - else: - nn.init.uniform_(vectors, -np.sqrt(3/dim), np.sqrt(3/dim)) - - if vocab._no_create_word_length>0: - if vocab.unknown is None: # 创建一个专门的unknown - unknown_idx = len(matrix) - vectors = torch.cat((vectors, torch.zeros(1, dim)), dim=0).contiguous() - else: - unknown_idx = vocab.unknown_idx - words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), - requires_grad=False) - for order, (index, vec) in enumerate(matrix.items()): - if vec is not None: - vectors[order] = vec - words_to_words[index] = order - self.words_to_words = words_to_words - else: - for index, vec in matrix.items(): - if vec is not None: - vectors[index] = vec - - if normalize: - vectors /= (torch.norm(vectors, dim=1, keepdim=True) + 1e-12) - - return vectors - - def forward(self, words): - """ - 传入words的index - - :param words: torch.LongTensor, [batch_size, max_len] - :return: torch.FloatTensor, [batch_size, max_len, embed_size] - """ - if hasattr(self, 'words_to_words'): - words = self.words_to_words[words] - words = self.drop_word(words) - words = self.embedding(words) - words = self.dropout(words) - return words - - -class ContextualEmbedding(TokenEmbedding): - def __init__(self, vocab: Vocabulary, word_dropout:float=0.0, dropout:float=0.0): - super(ContextualEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - def add_sentence_cache(self, *datasets, batch_size=32, device='cpu', delete_weights: bool=True): - """ - 由于动态embedding生成比较耗时，所以可以把每句话embedding缓存下来，这样就不需要每次都运行生成过程。 - - :param datasets: DataSet对象 - :param batch_size: int, 生成cache的sentence表示时使用的batch的大小 - :param device: 参考 :class::fastNLP.Trainer 的device - :param delete_weights: 似乎在生成了cache之后删除权重，在不需要finetune动态模型的情况下，删除权重会大量减少内存占用。 - :return: - """ - for index, dataset in enumerate(datasets): - try: - assert isinstance(dataset, DataSet), "Only fastNLP.DataSet object is allowed." - assert 'words' in dataset.get_input_name(), "`words` field has to be set as input." - except Exception as e: - print(f"Exception happens at {index} dataset.") - raise e - - sent_embeds = {} - _move_model_to_device(self, device=device) - device = _get_model_device(self) - pad_index = self._word_vocab.padding_idx - print("Start to calculate sentence representations.") - with torch.no_grad(): - for index, dataset in enumerate(datasets): - try: - batch = DataSetIter(dataset, batch_size=batch_size, sampler=SequentialSampler()) - for batch_x, batch_y in batch: - words = batch_x['words'].to(device) - words_list = words.tolist() - seq_len = words.ne(pad_index).sum(dim=-1) - max_len = words.size(1) - # 因为有些情况可能包含CLS, SEP, 从后面往前计算比较安全。 - seq_len_from_behind = (max_len - seq_len).tolist() - word_embeds = self(words).detach().cpu().numpy() - for b in range(words.size(0)): - length = seq_len_from_behind[b] - if length==0: - sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b] - else: - sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b, :-length] - except Exception as e: - print(f"Exception happens at {index} dataset.") - raise e - print("Finish calculating sentence representations.") - self.sent_embeds = sent_embeds - if delete_weights: - self._delete_model_weights() - - def _get_sent_reprs(self, words): - """ - 获取sentence的表示，如果有缓存，则返回缓存的值; 没有缓存则返回None - - :param words: torch.LongTensor - :return: - """ - if hasattr(self, 'sent_embeds'): - words_list = words.tolist() - seq_len = words.ne(self._word_pad_index).sum(dim=-1) - _embeds = [] - for b in range(len(words)): - words_i = tuple(words_list[b][:seq_len[b]]) - embed = self.sent_embeds[words_i] - _embeds.append(embed) - max_sent_len = max(map(len, _embeds)) - embeds = words.new_zeros(len(_embeds), max_sent_len, self.embed_size, dtype=torch.float, - device=words.device) - for i, embed in enumerate(_embeds): - embeds[i, :len(embed)] = torch.FloatTensor(embed).to(words.device) - return embeds - return None - - @abstractmethod - def _delete_model_weights(self): - """删除计算表示的模型以节省资源""" - raise NotImplementedError - - def remove_sentence_cache(self): - """ - 删除缓存的句子表示. 删除之后如果模型权重没有被删除，将开始使用动态计算权重。 - - :return: - """ - del self.sent_embeds - - -class ElmoEmbedding(ContextualEmbedding): - """ - 别名：:class:`fastNLP.modules.ElmoEmbedding` :class:`fastNLP.modules.encoder.embedding.ElmoEmbedding` - - 使用ELMo的embedding。初始化之后，只需要传入words就可以得到对应的embedding。 - 我们提供的ELMo预训练模型来自 https://github.com/HIT-SCIR/ELMoForManyLangs - - Example:: - - >>> embedding = ElmoEmbedding(vocab, model_dir_or_name='en', layers='2', requires_grad=True) - - :param vocab: 词表 - :param model_dir_or_name: 可以有两种方式调用预训练好的ELMo embedding：第一种是传入ELMo权重的文件名，第二种是传入ELMo版本的名称， - 目前支持的ELMo包括{`en` : 英文版本的ELMo, `cn` : 中文版本的ELMo,}。第二种情况将自动查看缓存中是否存在该模型，没有的话将自动下载 - :param layers: str, 指定返回的层数, 以,隔开不同的层。如果要返回第二层的结果'2', 返回后两层的结果'1,2'。不同的层的结果 - 按照这个顺序concat起来。默认为'2'。'mix'会使用可学习的权重结合不同层的表示(权重是否可训练与requires_grad保持一致， - 初始化权重对三层结果进行mean-pooling, 可以通过ElmoEmbedding.set_mix_weights_requires_grad()方法只将mix weights设置为可学习。) - :param requires_grad: bool, 该层是否需要gradient, 默认为False. - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 - :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 - :param cache_word_reprs: 可以选择对word的表示进行cache; 设置为True的话，将在初始化的时候为每个word生成对应的embedding， - 并删除character encoder，之后将直接使用cache的embedding。默认为False。 - """ - def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en', layers: str='2', requires_grad: bool=False, - word_dropout=0.0, dropout=0.0, cache_word_reprs: bool=False): - super(ElmoEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - # 根据model_dir_or_name检查是否存在并下载 - if model_dir_or_name.lower() in PRETRAINED_ELMO_MODEL_DIR: - PRETRAIN_URL = _get_base_url('elmo') - model_name = PRETRAINED_ELMO_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_dir = model_dir_or_name - else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") - self.model = _ElmoModel(model_dir, vocab, cache_word_reprs=cache_word_reprs) - - if layers=='mix': - self.layer_weights = nn.Parameter(torch.zeros(self.model.config['lstm']['n_layers']+1), - requires_grad=requires_grad) - self.gamma = nn.Parameter(torch.ones(1), requires_grad=requires_grad) - self._get_outputs = self._get_mixed_outputs - self._embed_size = self.model.config['lstm']['projection_dim'] * 2 - else: - layers = list(map(int, layers.split(','))) - assert len(layers) > 0, "Must choose one output" - for layer in layers: - assert 0 <= layer <= 2, "Layer index should be in range [0, 2]." - self.layers = layers - self._get_outputs = self._get_layer_outputs - self._embed_size = len(self.layers) * self.model.config['lstm']['projection_dim'] * 2 - - self.requires_grad = requires_grad - - def _get_mixed_outputs(self, outputs): - # outputs: num_layers x batch_size x max_len x hidden_size - # return: batch_size x max_len x hidden_size - weights = F.softmax(self.layer_weights+1/len(outputs), dim=0).to(outputs) - outputs = torch.einsum('l,lbij->bij', weights, outputs) - return self.gamma.to(outputs)*outputs - - def set_mix_weights_requires_grad(self, flag=True): - """ - 当初始化ElmoEmbedding时layers被设置为mix时，可以通过调用该方法设置mix weights是否可训练。如果layers不是mix，调用 - 该方法没有用。 - :param bool flag: 混合不同层表示的结果是否可以训练。 - :return: - """ - if hasattr(self, 'layer_weights'): - self.layer_weights.requires_grad = flag - self.gamma.requires_grad = flag - - def _get_layer_outputs(self, outputs): - if len(self.layers) == 1: - outputs = outputs[self.layers[0]] - else: - outputs = torch.cat(tuple([*outputs[self.layers]]), dim=-1) - - return outputs - - def forward(self, words: torch.LongTensor): - """ - 计算words的elmo embedding表示。根据elmo文章中介绍的ELMO实际上是有2L+1层结果，但是为了让结果比较容易拆分，token的 - 被重复了一次，使得实际上layer=0的结果是[token_embedding;token_embedding], 而layer=1的结果是[forward_hiddens; - backward_hiddens]. - - :param words: batch_size x max_len - :return: torch.FloatTensor. batch_size x max_len x (512*len(self.layers)) - """ - words = self.drop_word(words) - outputs = self._get_sent_reprs(words) - if outputs is not None: - return self.dropout(outputs) - outputs = self.model(words) - outputs = self._get_outputs(outputs) - return self.dropout(outputs) - - def _delete_model_weights(self): - for name in ['layers', 'model', 'layer_weights', 'gamma']: - if hasattr(self, name): - delattr(self, name) - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - - :return: - """ - requires_grads = set([param.requires_grad for name, param in self.named_parameters() - if 'words_to_chars_embedding' not in name and 'words_to_words' not in name]) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' in name or 'words_to_words' in name: # 这个不能加入到requires_grad中 - continue - param.requires_grad = value - - -class BertEmbedding(ContextualEmbedding): - """ - 别名：:class:`fastNLP.modules.BertEmbedding` :class:`fastNLP.modules.encoder.embedding.BertEmbedding` - - 使用BERT对words进行encode的Embedding。建议将输入的words长度限制在450以内，而不要使用512。这是由于预训练的bert模型长 - 度限制为512个token，而因为输入的word是未进行word piece分割的，在分割之后长度可能会超过最大长度限制。 - - Example:: - - >>> embedding = BertEmbedding(vocab, model_dir_or_name='en-base-uncased', requires_grad=False, layers='4,-2,-1') - - - :param fastNLP.Vocabulary vocab: 词表 - :param str model_dir_or_name: 模型所在目录或者模型的名称。默认值为 ``en-base-uncased``. - :param str layers:最终结果中的表示。以','隔开层数，可以以负数去索引倒数几层 - :param str pool_method: 因为在bert中，每个word会被表示为多个word pieces, 当获取一个word的表示的时候，怎样从它的word pieces - 中计算得到它对应的表示。支持``last``, ``first``, ``avg``, ``max``。 - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 - :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 - :param bool include_cls_sep: bool，在bert计算句子的表示的时候，需要在前面加上[CLS]和[SEP], 是否在结果中保留这两个内容。 这样 - 会使得word embedding的结果比输入的结果长两个token。在使用 :class::StackEmbedding 可能会遇到问题。 - :param bool requires_grad: 是否需要gradient。 - """ - def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en-base-uncased', layers: str='-1', - pool_method: str='first', word_dropout=0, dropout=0, requires_grad: bool=False, - include_cls_sep: bool=False): - super(BertEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - # 根据model_dir_or_name检查是否存在并下载 - if model_dir_or_name.lower() in PRETRAINED_BERT_MODEL_DIR: - PRETRAIN_URL = _get_base_url('bert') - model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_dir = model_dir_or_name - else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") - - self.model = _WordBertModel(model_dir=model_dir, vocab=vocab, layers=layers, - pool_method=pool_method, include_cls_sep=include_cls_sep) - - self.requires_grad = requires_grad - self._embed_size = len(self.model.layers)*self.model.encoder.hidden_size - - def _delete_model_weights(self): - del self.model - - def forward(self, words): - """ - 计算words的bert embedding表示。计算之前会在每句话的开始增加[CLS]在结束增加[SEP], 并根据include_cls_sep判断要不要 - 删除这两个token的表示。 - - :param torch.LongTensor words: [batch_size, max_len] - :return: torch.FloatTensor. batch_size x max_len x (768*len(self.layers)) - """ - words = self.drop_word(words) - outputs = self._get_sent_reprs(words) - if outputs is not None: - return self.dropout(words) - outputs = self.model(words) - outputs = torch.cat([*outputs], dim=-1) - - return self.dropout(words) - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - requires_grads = set([param.requires_grad for name, param in self.named_parameters() - if 'word_pieces_lengths' not in name]) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - if 'word_pieces_lengths' in name: # 这个不能加入到requires_grad中 - continue - param.requires_grad = value - - -def _construct_char_vocab_from_vocab(vocab:Vocabulary, min_freq:int=1): - """ - 给定一个word的vocabulary生成character的vocabulary. - - :param vocab: 从vocab - :param min_freq: - :return: - """ - char_vocab = Vocabulary(min_freq=min_freq) - for word, index in vocab: - if not vocab._is_word_no_create_entry(word): - char_vocab.add_word_lst(list(word)) - return char_vocab - - -class CNNCharEmbedding(TokenEmbedding): - """ - 别名：:class:`fastNLP.modules.CNNCharEmbedding` :class:`fastNLP.modules.encoder.embedding.CNNCharEmbedding` - - 使用CNN生成character embedding。CNN的结果为, embed(x) -> Dropout(x) -> CNN(x) -> activation(x) -> pool -> fc -> Dropout. - 不同的kernel大小的fitler结果是concat起来的。 - - Example:: - - >>> cnn_char_embed = CNNCharEmbedding(vocab) - - - :param vocab: 词表 - :param embed_size: 该word embedding的大小，默认值为50. - :param char_emb_size: character的embed的大小。character是从vocab中生成的。默认值为50. - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 - :param float dropout: 以多大的概率drop - :param filter_nums: filter的数量. 长度需要和kernels一致。默认值为[40, 30, 20]. - :param kernel_sizes: kernel的大小. 默认值为[5, 3, 1]. - :param pool_method: character的表示在合成一个表示时所使用的pool方法，支持'avg', 'max'. - :param activation: CNN之后使用的激活方法，支持'relu', 'sigmoid', 'tanh' 或者自定义函数. - :param min_char_freq: character的最少出现次数。默认值为2. - """ - def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, - dropout:float=0.5, filter_nums: List[int]=(40, 30, 20), kernel_sizes: List[int]=(5, 3, 1), - pool_method: str='max', activation='relu', min_char_freq: int=2): - super(CNNCharEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - for kernel in kernel_sizes: - assert kernel % 2 == 1, "Only odd kernel is allowed." - - assert pool_method in ('max', 'avg') - self.dropout = nn.Dropout(dropout) - self.pool_method = pool_method - # activation function - if isinstance(activation, str): - if activation.lower() == 'relu': - self.activation = F.relu - elif activation.lower() == 'sigmoid': - self.activation = F.sigmoid - elif activation.lower() == 'tanh': - self.activation = F.tanh - elif activation is None: - self.activation = lambda x: x - elif callable(activation): - self.activation = activation - else: - raise Exception( - "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") - - print("Start constructing character vocabulary.") - # 建立char的词表 - self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) - self.char_pad_index = self.char_vocab.padding_idx - print(f"In total, there are {len(self.char_vocab)} distinct characters.") - # 对vocab进行index - max_word_len = max(map(lambda x: len(x[0]), vocab)) - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), max_word_len), - fill_value=self.char_pad_index, dtype=torch.long), - requires_grad=False) - self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) - for word, index in vocab: - # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了。修改为不区分pad, 这样所有的也是同一个embed - self.words_to_chars_embedding[index, :len(word)] = \ - torch.LongTensor([self.char_vocab.to_index(c) for c in word]) - self.word_lengths[index] = len(word) - self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) - - self.convs = nn.ModuleList([nn.Conv1d( - char_emb_size, filter_nums[i], kernel_size=kernel_sizes[i], bias=True, padding=kernel_sizes[i] // 2) - for i in range(len(kernel_sizes))]) - self._embed_size = embed_size - self.fc = nn.Linear(sum(filter_nums), embed_size) - self.init_param() - - def forward(self, words): - """ - 输入words的index后，生成对应的words的表示。 - - :param words: [batch_size, max_len] - :return: [batch_size, max_len, embed_size] - """ - words = self.drop_word(words) - batch_size, max_len = words.size() - chars = self.words_to_chars_embedding[words] # batch_size x max_len x max_word_len - word_lengths = self.word_lengths[words] # batch_size x max_len - max_word_len = word_lengths.max() - chars = chars[:, :, :max_word_len] - # 为1的地方为mask - chars_masks = chars.eq(self.char_pad_index) # batch_size x max_len x max_word_len 如果为0, 说明是padding的位置了 - chars = self.char_embedding(chars) # batch_size x max_len x max_word_len x embed_size - chars = self.dropout(chars) - reshaped_chars = chars.reshape(batch_size*max_len, max_word_len, -1) - reshaped_chars = reshaped_chars.transpose(1, 2) # B' x E x M - conv_chars = [conv(reshaped_chars).transpose(1, 2).reshape(batch_size, max_len, max_word_len, -1) - for conv in self.convs] - conv_chars = torch.cat(conv_chars, dim=-1).contiguous() # B x max_len x max_word_len x sum(filters) - conv_chars = self.activation(conv_chars) - if self.pool_method == 'max': - conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) - chars, _ = torch.max(conv_chars, dim=-2) # batch_size x max_len x sum(filters) - else: - conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), 0) - chars = torch.sum(conv_chars, dim=-2)/chars_masks.eq(0).sum(dim=-1, keepdim=True).float() - chars = self.fc(chars) - return self.dropout(chars) - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - params = [] - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' not in name and 'word_lengths' not in name: - params.append(param.requires_grad) - requires_grads = set(params) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能加入到requires_grad中 - continue - param.requires_grad = value - - def init_param(self): - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能reset - continue - if param.data.dim()>1: - nn.init.xavier_uniform_(param, 1) - else: - nn.init.uniform_(param, -1, 1) - -class LSTMCharEmbedding(TokenEmbedding): - """ - 别名：:class:`fastNLP.modules.LSTMCharEmbedding` :class:`fastNLP.modules.encoder.embedding.LSTMCharEmbedding` - - 使用LSTM的方式对character进行encode. embed(x) -> Dropout(x) -> LSTM(x) -> activation(x) -> pool - - Example:: - - >>> lstm_char_embed = LSTMCharEmbedding(vocab) - - :param vocab: 词表 - :param embed_size: embedding的大小。默认值为50. - :param char_emb_size: character的embedding的大小。默认值为50. - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 - :param dropout: 以多大概率drop - :param hidden_size: LSTM的中间hidden的大小，如果为bidirectional的，hidden会除二，默认为50. - :param pool_method: 支持'max', 'avg' - :param activation: 激活函数，支持'relu', 'sigmoid', 'tanh', 或者自定义函数. - :param min_char_freq: character的最小出现次数。默认值为2. - :param bidirectional: 是否使用双向的LSTM进行encode。默认值为True。 - """ - def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, - dropout:float=0.5, hidden_size=50,pool_method: str='max', activation='relu', min_char_freq: int=2, - bidirectional=True): - super(LSTMCharEmbedding, self).__init__(vocab) - - assert hidden_size % 2 == 0, "Only even kernel is allowed." - - assert pool_method in ('max', 'avg') - self.pool_method = pool_method - self.dropout = nn.Dropout(dropout) - # activation function - if isinstance(activation, str): - if activation.lower() == 'relu': - self.activation = F.relu - elif activation.lower() == 'sigmoid': - self.activation = F.sigmoid - elif activation.lower() == 'tanh': - self.activation = F.tanh - elif activation is None: - self.activation = lambda x: x - elif callable(activation): - self.activation = activation - else: - raise Exception( - "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") - - print("Start constructing character vocabulary.") - # 建立char的词表 - self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) - self.char_pad_index = self.char_vocab.padding_idx - print(f"In total, there are {len(self.char_vocab)} distinct characters.") - # 对vocab进行index - self.max_word_len = max(map(lambda x: len(x[0]), vocab)) - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), self.max_word_len), - fill_value=self.char_pad_index, dtype=torch.long), - requires_grad=False) - self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) - for word, index in vocab: - # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了. 修改为不区分pad与否 - self.words_to_chars_embedding[index, :len(word)] = \ - torch.LongTensor([self.char_vocab.to_index(c) for c in word]) - self.word_lengths[index] = len(word) - self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) - - self.fc = nn.Linear(hidden_size, embed_size) - hidden_size = hidden_size // 2 if bidirectional else hidden_size - - self.lstm = LSTM(char_emb_size, hidden_size, bidirectional=bidirectional, batch_first=True) - self._embed_size = embed_size - self.bidirectional = bidirectional - - def forward(self, words): - """ - 输入words的index后，生成对应的words的表示。 - - :param words: [batch_size, max_len] - :return: [batch_size, max_len, embed_size] - """ - words = self.drop_word(words) - batch_size, max_len = words.size() - chars = self.words_to_chars_embedding[words] # batch_size x max_len x max_word_len - word_lengths = self.word_lengths[words] # batch_size x max_len - max_word_len = word_lengths.max() - chars = chars[:, :, :max_word_len] - # 为mask的地方为1 - chars_masks = chars.eq(self.char_pad_index) # batch_size x max_len x max_word_len 如果为0, 说明是padding的位置了 - chars = self.char_embedding(chars) # batch_size x max_len x max_word_len x embed_size - chars = self.dropout(chars) - reshaped_chars = chars.reshape(batch_size * max_len, max_word_len, -1) - char_seq_len = chars_masks.eq(0).sum(dim=-1).reshape(batch_size * max_len) - lstm_chars = self.lstm(reshaped_chars, char_seq_len)[0].reshape(batch_size, max_len, max_word_len, -1) - # B x M x M x H - - lstm_chars = self.activation(lstm_chars) - if self.pool_method == 'max': - lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) - chars, _ = torch.max(lstm_chars, dim=-2) # batch_size x max_len x H - else: - lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), 0) - chars = torch.sum(lstm_chars, dim=-2) / chars_masks.eq(0).sum(dim=-1, keepdim=True).float() - - chars = self.fc(chars) - - return self.dropout(chars) - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - params = [] - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' not in name and 'word_lengths' not in name: - params.append(param) - requires_grads = set(params) - if len(requires_grads) == 1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for name, param in self.named_parameters(): - if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能加入到requires_grad中 - continue - param.requires_grad = value - - -class StackEmbedding(TokenEmbedding): - """ - 别名：:class:`fastNLP.modules.StackEmbedding` :class:`fastNLP.modules.encoder.embedding.StackEmbedding` - - 支持将多个embedding集合成一个embedding。 - - Example:: - - >>> embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) - >>> embed_2 = StaticEmbedding(vocab, model_dir_or_name='en-word2vec-300', requires_grad=True) - - - :param embeds: 一个由若干个TokenEmbedding组成的list，要求每一个TokenEmbedding的词表都保持一致 - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。不同embedidng会在相同的位置 - 被设置为unknown。如果这里设置了dropout，则组成的embedding就不要再设置dropout了。 - :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 - - """ - def __init__(self, embeds: List[TokenEmbedding], word_dropout=0, dropout=0): - vocabs = [] - for embed in embeds: - if hasattr(embed, 'get_word_vocab'): - vocabs.append(embed.get_word_vocab()) - _vocab = vocabs[0] - for vocab in vocabs[1:]: - assert vocab == _vocab, "All embeddings in StackEmbedding should use the same word vocabulary." - - super(StackEmbedding, self).__init__(_vocab, word_dropout=word_dropout, dropout=dropout) - assert isinstance(embeds, list) - for embed in embeds: - assert isinstance(embed, TokenEmbedding), "Only TokenEmbedding type is supported." - self.embeds = nn.ModuleList(embeds) - self._embed_size = sum([embed.embed_size for embed in self.embeds]) - - def append(self, embed: TokenEmbedding): - """ - 添加一个embedding到结尾。 - :param embed: - :return: - """ - assert isinstance(embed, TokenEmbedding) - self.embeds.append(embed) - - def pop(self): - """ - 弹出最后一个embed - :return: - """ - return self.embeds.pop() - - @property - def embed_size(self): - return self._embed_size - - @property - def requires_grad(self): - """ - Embedding的参数是否允许优化。True: 所有参数运行优化; False: 所有参数不允许优化; None: 部分允许优化、部分不允许 - :return: - """ - requires_grads = set([embed.requires_grad for embed in self.embeds()]) - if len(requires_grads)==1: - return requires_grads.pop() - else: - return None - - @requires_grad.setter - def requires_grad(self, value): - for embed in self.embeds(): - embed.requires_grad = value - - def forward(self, words): - """ - 得到多个embedding的结果，并把结果按照顺序concat起来。 - - :param words: batch_size x max_len - :return: 返回的shape和当前这个stack embedding中embedding的组成有关 - """ - outputs = [] - words = self.drop_word(words) - for embed in self.embeds: - outputs.append(embed(words)) - outputs = self.dropout(torch.cat(outputs, dim=-1)) - return outputs - diff --git a/fastNLP/modules/encoder/lstm.py b/fastNLP/modules/encoder/lstm.py index 5e599a65..e2358132 100644 --- a/fastNLP/modules/encoder/lstm.py +++ b/fastNLP/modules/encoder/lstm.py @@ -10,13 +10,10 @@ import torch import torch.nn as nn import torch.nn.utils.rnn as rnn -from ..utils import initial_parameter -from torch import autograd - class LSTM(nn.Module): """ - 别名：:class:`fastNLP.modules.LSTM` :class:`fastNLP.modules.encoder.lstm.LSTM` + 别名：:class:`fastNLP.modules.LSTM` :class:`fastNLP.modules.encoder.LSTM` LSTM 模块, 轻量封装的Pytorch LSTM. 在提供seq_len的情况下，将自动使用pack_padded_sequence; 同时默认将forget gate的bias初始化 为1; 且可以应对DataParallel中LSTM的使用问题。 @@ -30,7 +27,7 @@ class LSTM(nn.Module): :(batch, seq, feature). Default: ``False`` :param bias: 如果为 ``False``, 模型将不会使用bias. Default: ``True`` """ - + def __init__(self, input_size, hidden_size=100, num_layers=1, dropout=0.0, batch_first=True, bidirectional=False, bias=True): super(LSTM, self).__init__() diff --git a/fastNLP/modules/encoder/pooling.py b/fastNLP/modules/encoder/pooling.py index 8337fe32..d8aa54ad 100644 --- a/fastNLP/modules/encoder/pooling.py +++ b/fastNLP/modules/encoder/pooling.py @@ -10,7 +10,7 @@ import torch.nn as nn class MaxPool(nn.Module): """ - 别名：:class:`fastNLP.modules.MaxPool` :class:`fastNLP.modules.encoder.pooling.MaxPool` + 别名：:class:`fastNLP.modules.MaxPool` :class:`fastNLP.modules.encoder.MaxPool` Max-pooling模块。 @@ -21,9 +21,9 @@ class MaxPool(nn.Module): :param kernel_size: max pooling的窗口大小，默认为tensor最后k维，其中k为dimension :param ceil_mode: """ - + def __init__(self, stride=None, padding=0, dilation=1, dimension=1, kernel_size=None, ceil_mode=False): - + super(MaxPool, self).__init__() assert (1 <= dimension) and (dimension <= 3) self.dimension = dimension @@ -32,7 +32,7 @@ class MaxPool(nn.Module): self.dilation = dilation self.kernel_size = kernel_size self.ceil_mode = ceil_mode - + def forward(self, x): if self.dimension == 1: pooling = nn.MaxPool1d( @@ -59,15 +59,15 @@ class MaxPool(nn.Module): class MaxPoolWithMask(nn.Module): """ - 别名：:class:`fastNLP.modules.MaxPoolWithMask` :class:`fastNLP.modules.encoder.pooling.MaxPoolWithMask` + 别名：:class:`fastNLP.modules.MaxPoolWithMask` :class:`fastNLP.modules.encoder.MaxPoolWithMask` 带mask矩阵的max pooling。在做max-pooling的时候不会考虑mask值为0的位置。 """ - + def __init__(self): super(MaxPoolWithMask, self).__init__() self.inf = 10e12 - + def forward(self, tensor, mask, dim=1): """ :param torch.FloatTensor tensor: [batch_size, seq_len, channels] 初始tensor @@ -82,11 +82,11 @@ class MaxPoolWithMask(nn.Module): class KMaxPool(nn.Module): """K max-pooling module.""" - + def __init__(self, k=1): super(KMaxPool, self).__init__() self.k = k - + def forward(self, x): """ :param torch.Tensor x: [N, C, L] 初始tensor @@ -99,16 +99,16 @@ class KMaxPool(nn.Module): class AvgPool(nn.Module): """ - 别名：:class:`fastNLP.modules.AvgPool` :class:`fastNLP.modules.encoder.pooling.AvgPool` + 别名：:class:`fastNLP.modules.AvgPool` :class:`fastNLP.modules.encoder.AvgPool` 给定形如[batch_size, max_len, hidden_size]的输入，在最后一维进行avg pooling. 输出为[batch_size, hidden_size] """ - + def __init__(self, stride=None, padding=0): super(AvgPool, self).__init__() self.stride = stride self.padding = padding - + def forward(self, x): """ :param torch.Tensor x: [N, C, L] 初始tensor @@ -126,16 +126,16 @@ class AvgPool(nn.Module): class AvgPoolWithMask(nn.Module): """ - 别名：:class:`fastNLP.modules.AvgPoolWithMask` :class:`fastNLP.modules.encoder.pooling.AvgPoolWithMask` + 别名：:class:`fastNLP.modules.AvgPoolWithMask` :class:`fastNLP.modules.encoder.AvgPoolWithMask` 给定形如[batch_size, max_len, hidden_size]的输入，在最后一维进行avg pooling. 输出为[batch_size, hidden_size], pooling 的时候只会考虑mask为1的位置 """ - + def __init__(self): super(AvgPoolWithMask, self).__init__() self.inf = 10e12 - + def forward(self, tensor, mask, dim=1): """ :param torch.FloatTensor tensor: [batch_size, seq_len, channels] 初始tensor diff --git a/fastNLP/modules/encoder/star_transformer.py b/fastNLP/modules/encoder/star_transformer.py index 097fbebb..3927a494 100644 --- a/fastNLP/modules/encoder/star_transformer.py +++ b/fastNLP/modules/encoder/star_transformer.py @@ -13,7 +13,7 @@ from torch.nn import functional as F class StarTransformer(nn.Module): """ - 别名：:class:`fastNLP.modules.StarTransformer` :class:`fastNLP.modules.encoder.star_transformer.StarTransformer` + 别名：:class:`fastNLP.modules.StarTransformer` :class:`fastNLP.modules.encoder.StarTransformer` Star-Transformer 的encoder部分。 输入3d的文本输入, 返回相同长度的文本编码 @@ -29,11 +29,11 @@ class StarTransformer(nn.Module): 模型会为输入序列加上position embedding。 若为`None`，忽略加上position embedding的步骤. Default: `None` """ - + def __init__(self, hidden_size, num_layers, num_head, head_dim, dropout=0.1, max_len=None): super(StarTransformer, self).__init__() self.iters = num_layers - + self.norm = nn.ModuleList([nn.LayerNorm(hidden_size, eps=1e-6) for _ in range(self.iters)]) # self.emb_fc = nn.Conv2d(hidden_size, hidden_size, 1) self.emb_drop = nn.Dropout(dropout) @@ -43,12 +43,12 @@ class StarTransformer(nn.Module): self.star_att = nn.ModuleList( [_MSA2(hidden_size, nhead=num_head, head_dim=head_dim, dropout=0.0) for _ in range(self.iters)]) - + if max_len is not None: self.pos_emb = nn.Embedding(max_len, hidden_size) else: self.pos_emb = None - + def forward(self, data, mask): """ :param FloatTensor data: [batch, length, hidden] 输入的序列 @@ -58,15 +58,15 @@ class StarTransformer(nn.Module): [batch, hidden] 全局 relay 节点, 详见论文 """ - + def norm_func(f, x): # B, H, L, 1 return f(x.permute(0, 2, 3, 1)).permute(0, 3, 1, 2) - + B, L, H = data.size() mask = (mask == 0) # flip the mask for masked_fill_ smask = torch.cat([torch.zeros(B, 1, ).byte().to(mask), mask], 1) - + embs = data.permute(0, 2, 1)[:, :, :, None] # B H L 1 if self.pos_emb and False: P = self.pos_emb(torch.arange(L, dtype=torch.long, device=embs.device) \ @@ -80,13 +80,13 @@ class StarTransformer(nn.Module): for i in range(self.iters): ax = torch.cat([r_embs, relay.expand(B, H, 1, L)], 2) nodes = F.leaky_relu(self.ring_att[i](norm_func(self.norm[i], nodes), ax=ax)) - #nodes = F.leaky_relu(self.ring_att[i](nodes, ax=ax)) + # nodes = F.leaky_relu(self.ring_att[i](nodes, ax=ax)) relay = F.leaky_relu(self.star_att[i](relay, torch.cat([relay, nodes], 2), smask)) - + nodes = nodes.masked_fill_(ex_mask, 0) - + nodes = nodes.view(B, H, L).permute(0, 2, 1) - + return nodes, relay.view(B, H) @@ -99,19 +99,19 @@ class _MSA1(nn.Module): self.WK = nn.Conv2d(nhid, nhead * head_dim, 1) self.WV = nn.Conv2d(nhid, nhead * head_dim, 1) self.WO = nn.Conv2d(nhead * head_dim, nhid, 1) - + self.drop = nn.Dropout(dropout) - + # print('NUM_HEAD', nhead, 'DIM_HEAD', head_dim) self.nhid, self.nhead, self.head_dim, self.unfold_size = nhid, nhead, head_dim, 3 - + def forward(self, x, ax=None): # x: B, H, L, 1, ax : B, H, X, L append features nhid, nhead, head_dim, unfold_size = self.nhid, self.nhead, self.head_dim, self.unfold_size B, H, L, _ = x.shape - + q, k, v = self.WQ(x), self.WK(x), self.WV(x) # x: (B,H,L,1) - + if ax is not None: aL = ax.shape[2] ak = self.WK(ax).view(B, nhead, head_dim, aL, L) @@ -124,12 +124,12 @@ class _MSA1(nn.Module): if ax is not None: k = torch.cat([k, ak], 3) v = torch.cat([v, av], 3) - + alphas = self.drop(F.softmax((q * k).sum(2, keepdim=True) / NP.sqrt(head_dim), 3)) # B N L 1 U att = (alphas * v).sum(3).view(B, nhead * head_dim, L, 1) - + ret = self.WO(att) - + return ret @@ -141,19 +141,19 @@ class _MSA2(nn.Module): self.WK = nn.Conv2d(nhid, nhead * head_dim, 1) self.WV = nn.Conv2d(nhid, nhead * head_dim, 1) self.WO = nn.Conv2d(nhead * head_dim, nhid, 1) - + self.drop = nn.Dropout(dropout) - + # print('NUM_HEAD', nhead, 'DIM_HEAD', head_dim) self.nhid, self.nhead, self.head_dim, self.unfold_size = nhid, nhead, head_dim, 3 - + def forward(self, x, y, mask=None): # x: B, H, 1, 1, 1 y: B H L 1 nhid, nhead, head_dim, unfold_size = self.nhid, self.nhead, self.head_dim, self.unfold_size B, H, L, _ = y.shape - + q, k, v = self.WQ(x), self.WK(y), self.WV(y) - + q = q.view(B, nhead, 1, head_dim) # B, H, 1, 1 -> B, N, 1, h k = k.view(B, nhead, head_dim, L) # B, H, L, 1 -> B, N, h, L v = v.view(B, nhead, head_dim, L).permute(0, 1, 3, 2) # B, H, L, 1 -> B, N, L, h diff --git a/fastNLP/modules/encoder/transformer.py b/fastNLP/modules/encoder/transformer.py index d6bf2f1e..bc488e54 100644 --- a/fastNLP/modules/encoder/transformer.py +++ b/fastNLP/modules/encoder/transformer.py @@ -9,7 +9,7 @@ from ..dropout import TimestepDropout class TransformerEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.TransformerEncoder` :class:`fastNLP.modules.encoder.transformer.TransformerEncoder` + 别名：:class:`fastNLP.modules.TransformerEncoder` :class:`fastNLP.modules.encoder.TransformerEncoder` transformer的encoder模块，不包含embedding层 @@ -22,7 +22,7 @@ class TransformerEncoder(nn.Module): :param int num_head: head的数量。 :param float dropout: dropout概率. Default: 0.1 """ - + class SubLayer(nn.Module): def __init__(self, model_size, inner_size, key_size, value_size, num_head, dropout=0.1): super(TransformerEncoder.SubLayer, self).__init__() @@ -33,7 +33,7 @@ class TransformerEncoder(nn.Module): nn.Linear(inner_size, model_size), TimestepDropout(dropout), ) self.norm2 = nn.LayerNorm(model_size) - + def forward(self, input, seq_mask=None, atte_mask_out=None): """ @@ -48,11 +48,11 @@ class TransformerEncoder(nn.Module): output = self.norm2(output + norm_atte) output *= seq_mask return output - + def __init__(self, num_layers, **kargs): super(TransformerEncoder, self).__init__() self.layers = nn.ModuleList([self.SubLayer(**kargs) for _ in range(num_layers)]) - + def forward(self, x, seq_mask=None): """ :param x: [batch, seq_len, model_size] 输入序列 diff --git a/fastNLP/modules/encoder/variational_rnn.py b/fastNLP/modules/encoder/variational_rnn.py index 29b728e5..8e5e804b 100644 --- a/fastNLP/modules/encoder/variational_rnn.py +++ b/fastNLP/modules/encoder/variational_rnn.py @@ -28,14 +28,14 @@ class VarRnnCellWrapper(nn.Module): """ Wrapper for normal RNN Cells, make it support variational dropout """ - + def __init__(self, cell, hidden_size, input_p, hidden_p): super(VarRnnCellWrapper, self).__init__() self.cell = cell self.hidden_size = hidden_size self.input_p = input_p self.hidden_p = hidden_p - + def forward(self, input_x, hidden, mask_x, mask_h, is_reversed=False): """ :param PackedSequence input_x: [seq_len, batch_size, input_size] @@ -47,13 +47,13 @@ class VarRnnCellWrapper(nn.Module): hidden: for LSTM, tuple of (h_n, c_n), [batch_size, hidden_size] for other RNN, h_n, [batch_size, hidden_size] """ - + def get_hi(hi, h0, size): h0_size = size - hi.size(0) if h0_size > 0: return torch.cat([hi, h0[:h0_size]], dim=0) return hi[:size] - + is_lstm = isinstance(hidden, tuple) input, batch_sizes = input_x.data, input_x.batch_sizes output = [] @@ -64,7 +64,7 @@ class VarRnnCellWrapper(nn.Module): else: batch_iter = batch_sizes idx = 0 - + if is_lstm: hn = (hidden[0].clone(), hidden[1].clone()) else: @@ -91,7 +91,7 @@ class VarRnnCellWrapper(nn.Module): hi = cell(input_i, hi) hn[:size] = hi output.append(hi) - + if is_reversed: output = list(reversed(output)) output = torch.cat(output, dim=0) @@ -117,7 +117,7 @@ class VarRNNBase(nn.Module): :param hidden_dropout: 对每个隐状态的dropout概率. Default: 0 :param bidirectional: 若为 ``True``, 使用双向的RNN. Default: ``False`` """ - + def __init__(self, mode, Cell, input_size, hidden_size, num_layers=1, bias=True, batch_first=False, input_dropout=0, hidden_dropout=0, bidirectional=False): @@ -141,7 +141,7 @@ class VarRNNBase(nn.Module): cell, self.hidden_size, input_dropout, hidden_dropout)) initial_parameter(self) self.is_lstm = (self.mode == "LSTM") - + def _forward_one(self, n_layer, n_direction, input, hx, mask_x, mask_h): is_lstm = self.is_lstm idx = self.num_directions * n_layer + n_direction @@ -150,7 +150,7 @@ class VarRNNBase(nn.Module): output_x, hidden_x = cell( input, hi, mask_x, mask_h, is_reversed=(n_direction == 1)) return output_x, hidden_x - + def forward(self, x, hx=None): """ @@ -170,13 +170,13 @@ class VarRNNBase(nn.Module): else: max_batch_size = int(x.batch_sizes[0]) x, batch_sizes = x.data, x.batch_sizes - + if hx is None: hx = x.new_zeros(self.num_layers * self.num_directions, max_batch_size, self.hidden_size, requires_grad=True) if is_lstm: hx = (hx, hx.new_zeros(hx.size(), requires_grad=True)) - + mask_x = x.new_ones((max_batch_size, self.input_size)) mask_out = x.new_ones( (max_batch_size, self.hidden_size * self.num_directions)) @@ -185,7 +185,7 @@ class VarRNNBase(nn.Module): training=self.training, inplace=True) nn.functional.dropout(mask_out, p=self.hidden_dropout, training=self.training, inplace=True) - + hidden = x.new_zeros( (self.num_layers * self.num_directions, max_batch_size, self.hidden_size)) if is_lstm: @@ -207,22 +207,22 @@ class VarRNNBase(nn.Module): else: hidden[idx] = hidden_x x = torch.cat(output_list, dim=-1) - + if is_lstm: hidden = (hidden, cellstate) - + if is_packed: output = PackedSequence(x, batch_sizes) else: x = PackedSequence(x, batch_sizes) output, _ = pad_packed_sequence(x, batch_first=self.batch_first) - + return output, hidden class VarLSTM(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarLSTM` :class:`fastNLP.modules.encoder.variational_rnn.VarLSTM` + 别名：:class:`fastNLP.modules.VarLSTM` :class:`fastNLP.modules.encoder.VarLSTM` Variational Dropout LSTM. @@ -236,18 +236,18 @@ class VarLSTM(VarRNNBase): :param hidden_dropout: 对每个隐状态的dropout概率. Default: 0 :param bidirectional: 若为 ``True``, 使用双向的LSTM. Default: ``False`` """ - + def __init__(self, *args, **kwargs): super(VarLSTM, self).__init__( mode="LSTM", Cell=nn.LSTMCell, *args, **kwargs) - + def forward(self, x, hx=None): return super(VarLSTM, self).forward(x, hx) class VarRNN(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarRNN` :class:`fastNLP.modules.encoder.variational_rnn.VarRNN` + 别名：:class:`fastNLP.modules.VarRNN` :class:`fastNLP.modules.encoder.VarRNN` Variational Dropout RNN. @@ -261,18 +261,18 @@ class VarRNN(VarRNNBase): :param hidden_dropout: 对每个隐状态的dropout概率. Default: 0 :param bidirectional: 若为 ``True``, 使用双向的RNN. Default: ``False`` """ - + def __init__(self, *args, **kwargs): super(VarRNN, self).__init__( mode="RNN", Cell=nn.RNNCell, *args, **kwargs) - + def forward(self, x, hx=None): return super(VarRNN, self).forward(x, hx) class VarGRU(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarGRU` :class:`fastNLP.modules.encoder.variational_rnn.VarGRU` + 别名：:class:`fastNLP.modules.VarGRU` :class:`fastNLP.modules.encoder.VarGRU` Variational Dropout GRU. @@ -286,10 +286,10 @@ class VarGRU(VarRNNBase): :param hidden_dropout: 对每个隐状态的dropout概率. Default: 0 :param bidirectional: 若为 ``True``, 使用双向的GRU. Default: ``False`` """ - + def __init__(self, *args, **kwargs): super(VarGRU, self).__init__( mode="GRU", Cell=nn.GRUCell, *args, **kwargs) - + def forward(self, x, hx=None): return super(VarGRU, self).forward(x, hx) diff --git a/fastNLP/modules/utils.py b/fastNLP/modules/utils.py index 3c6a3d27..21608c5d 100644 --- a/fastNLP/modules/utils.py +++ b/fastNLP/modules/utils.py @@ -1,10 +1,10 @@ from functools import reduce -import numpy as np import torch import torch.nn as nn import torch.nn.init as init - +import glob +import os def initial_parameter(net, initial_method=None): """A method used to initialize the weights of PyTorch models. @@ -70,33 +70,6 @@ def initial_parameter(net, initial_method=None): net.apply(weights_init) -def get_embeddings(init_embed): - """ - 根据输入的init_embed生成nn.Embedding对象。 - - :param init_embed: 可以是 tuple:(num_embedings, embedding_dim), 即embedding的大小和每个词的维度;也可以传入 - nn.Embedding 对象, 此时就以传入的对象作为embedding; 传入np.ndarray也行，将使用传入的ndarray作为作为Embedding初始 - 化; 传入orch.Tensor, 将使用传入的值作为Embedding初始化。 - :return nn.Embedding embeddings: - """ - if isinstance(init_embed, tuple): - res = nn.Embedding( - num_embeddings=init_embed[0], embedding_dim=init_embed[1]) - nn.init.uniform_(res.weight.data, a=-np.sqrt(3/res.weight.data.size(1)), - b=np.sqrt(3/res.weight.data.size(1))) - elif isinstance(init_embed, nn.Module): - res = init_embed - elif isinstance(init_embed, torch.Tensor): - res = nn.Embedding.from_pretrained(init_embed, freeze=False) - elif isinstance(init_embed, np.ndarray): - init_embed = torch.tensor(init_embed, dtype=torch.float32) - res = nn.Embedding.from_pretrained(init_embed, freeze=False) - else: - raise TypeError( - 'invalid init_embed type: {}'.format((type(init_embed)))) - return res - - def summary(model: nn.Module): """ 得到模型的总参数量 @@ -145,4 +118,19 @@ def get_dropout_mask(drop_p: float, tensor: torch.Tensor): mask_x = torch.ones_like(tensor) nn.functional.dropout(mask_x, p=drop_p, training=False, inplace=True) - return mask_x \ No newline at end of file + return mask_x + + +def _get_file_name_base_on_postfix(dir_path, postfix): + """ + 在dir_path中寻找后缀为postfix的文件. + :param dir_path: str, 文件夹 + :param postfix: 形如".bin", ".json"等 + :return: str，文件的路径 + """ + files = list(filter(lambda filename:filename.endswith(postfix), os.listdir(os.path.join(dir_path)))) + if len(files) == 0: + raise FileNotFoundError(f"There is no file endswith *{postfix} file in {dir_path}") + elif len(files) > 1: + raise FileExistsError(f"There are multiple *{postfix} files in {dir_path}") + return os.path.join(dir_path, files[0]) \ No newline at end of file diff --git a/reproduction/Star_transformer/train.py b/reproduction/Star_transformer/train.py index f1e5c2f9..d8e2576b 100644 --- a/reproduction/Star_transformer/train.py +++ b/reproduction/Star_transformer/train.py @@ -1,7 +1,7 @@ -from util import get_argparser, set_gpu, set_rng_seeds, add_model_args +from reproduction.Star_transformer.util import get_argparser, set_gpu, set_rng_seeds, add_model_args seed = set_rng_seeds(15360) print('RNG SEED {}'.format(seed)) -from datasets import load_seqtag, load_sst, load_snli, EmbedLoader, MAX_LEN +from reproduction.Star_transformer.datasets import load_seqtag, load_sst, load_snli, EmbedLoader, MAX_LEN import torch.nn as nn import torch import numpy as np diff --git a/reproduction/Summarization/Baseline/data/dataloader.py b/reproduction/Summarization/Baseline/data/dataloader.py index 57702904..47cd0856 100644 --- a/reproduction/Summarization/Baseline/data/dataloader.py +++ b/reproduction/Summarization/Baseline/data/dataloader.py @@ -56,7 +56,7 @@ class SummarizationLoader(JsonLoader): return ds - def process(self, paths, vocab_size, vocab_path, sent_max_len, doc_max_timesteps, domain=False, tag=False, load_vocab=True): + def process(self, paths, vocab_size, vocab_path, sent_max_len, doc_max_timesteps, domain=False, tag=False, load_vocab_file=True): """ :param paths: dict path for each dataset :param vocab_size: int max_size for vocab @@ -65,7 +65,7 @@ class SummarizationLoader(JsonLoader): :param doc_max_timesteps: int max sentence number of the document :param domain: bool build vocab for publication, use 'X' for unknown :param tag: bool build vocab for tag, use 'X' for unknown - :param load_vocab: bool build vocab (False) or load vocab (True) + :param load_vocab_file: bool build vocab (False) or load vocab (True) :return: DataBundle datasets: dict keys correspond to the paths dict vocabs: dict key: vocab(if "train" in paths), domain(if domain=True), tag(if tag=True) @@ -146,7 +146,7 @@ class SummarizationLoader(JsonLoader): train_ds = datasets[key] vocab_dict = {} - if load_vocab == False: + if load_vocab_file == False: logger.info("[INFO] Build new vocab from training dataset!") if train_ds == None: raise ValueError("Lack train file to build vocabulary!") diff --git a/reproduction/Summarization/Baseline/tools/data.py b/reproduction/Summarization/Baseline/tools/data.py index f7bbaddd..0cbfbb06 100644 --- a/reproduction/Summarization/Baseline/tools/data.py +++ b/reproduction/Summarization/Baseline/tools/data.py @@ -36,8 +36,8 @@ import pickle from nltk.tokenize import sent_tokenize -import utils -from logger import * +import tools.utils +from tools.logger import * # and are used in the data files to segment the abstracts into sentences. They don't receive vocab ids. SENTENCE_START = '' @@ -313,7 +313,8 @@ class Example(object): for sent in article_sents: article_words = sent.split() self.enc_sent_len.append(len(article_words)) # store the length after truncation but before padding - self.enc_sent_input.append([vocab.word2id(w) for w in article_words]) # list of word ids; OOVs are represented by the id for UNK token + # self.enc_sent_input.append([vocab.word2id(w) for w in article_words]) # list of word ids; OOVs are represented by the id for UNK token + self.enc_sent_input.append([vocab.word2id(w.lower()) for w in article_words]) # list of word ids; OOVs are represented by the id for UNK token self._pad_encoder_input(vocab.word2id('[PAD]')) # Store the original strings diff --git a/reproduction/Summarization/Baseline/train.py b/reproduction/Summarization/Baseline/train.py index c3a92f67..b3170307 100644 --- a/reproduction/Summarization/Baseline/train.py +++ b/reproduction/Summarization/Baseline/train.py @@ -29,7 +29,7 @@ import torch.nn os.environ['FASTNLP_BASE_URL'] = 'http://10.141.222.118:8888/file/download/' os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' -sys.path.append('/remote-home/dqwang/FastNLP/fastNLP/') +sys.path.append('/remote-home/dqwang/FastNLP/fastNLP_brxx/') from fastNLP.core.const import Const diff --git a/reproduction/Summarization/BertSum/model.py b/reproduction/Summarization/BertSum/model.py index 655ad16e..1ee821fc 100644 --- a/reproduction/Summarization/BertSum/model.py +++ b/reproduction/Summarization/BertSum/model.py @@ -2,7 +2,7 @@ import torch from torch import nn from torch.nn import init -from fastNLP.modules.encoder._bert import BertModel +from fastNLP.modules.encoder.bert import BertModel class Classifier(nn.Module): diff --git a/reproduction/Summarization/README.md b/reproduction/Summarization/README.md index 2cde6304..b584269f 100644 --- a/reproduction/Summarization/README.md +++ b/reproduction/Summarization/README.md @@ -39,6 +39,54 @@ FastNLP中实现的模型包括： +### Evaluation + +#### FastRougeMetric + +FastRougeMetric使用python实现的ROUGE非官方库来实现在训练过程中快速计算rouge近似值。 + 源代码可见 [https://github.com/pltrdy/rouge](https://github.com/pltrdy/rouge) + +在fastNLP中，该方法已经被包装成Metric.py中的FastRougeMetric类以供trainer直接使用。 +需要事先使用pip安装该rouge库。 + + pip install rouge + + +**注意：由于实现细节的差异，该结果和官方ROUGE结果存在1-2个点的差异，仅可作为训练过程优化趋势的粗略估计。** + + + +#### PyRougeMetric + +PyRougeMetric 使用论文 [*ROUGE: A Package for Automatic Evaluation of Summaries*](https://www.aclweb.org/anthology/W04-1013) 提供的官方ROUGE 1.5.5评测库。 + +由于原本的ROUGE使用perl解释器，[pyrouge](https://github.com/bheinzerling/pyrouge)对其进行了python包装，而PyRougeMetric将其进一步包装为trainer可以直接使用的Metric类。 + +为了使用ROUGE 1.5.5，需要使用sudo权限安装一系列依赖库。 + +1. ROUGE 本身在Ubuntu下的安装可以参考[博客](https://blog.csdn.net/Hay54/article/details/78744912) +2. 配置wordnet可参考： +```shell +$ cd ~/rouge/RELEASE-1.5.5/data/WordNet-2.0-Exceptions/ +$ ./buildExeptionDB.pl . exc WordNet-2.0.exc.db +$ cd ../ +$ ln -s WordNet-2.0-Exceptions/WordNet-2.0.exc.db WordNet-2.0.exc.db +``` +3. 安装pyrouge +```shell +$ git clone https://github.com/bheinzerling/pyrouge +$ cd pyrouge +$ python setup.py install +``` +4. 测试ROUGE安装是否正确 +```shell +$ pyrouge_set_rouge_path /absolute/path/to/ROUGE-1.5.5/directory +$ python -m pyrouge.test +``` + + + + ### Dataset_loader - SummarizationLoader: 用于读取处理好的jsonl格式数据集，返回以下field @@ -56,6 +104,21 @@ FastNLP中实现的模型包括： +### Train Cmdline + +#### Baseline + +LSTM + Sequence Labeling + + python train.py --cuda --gpu --sentence_encoder deeplstm --sentence_decoder seqlab --save_root --log_root --lr_descent --grad_clip --max_grad_norm 10 + +Transformer + Sequence Labeling + + python train.py --cuda --gpu --sentence_encoder transformer --sentence_decoder seqlab --save_root --log_root --lr_descent --grad_clip --max_grad_norm 10 + + + +#### BertSum diff --git a/reproduction/joint_cws_parse/readme.md b/reproduction/joint_cws_parse/README.md similarity index 100% rename from reproduction/joint_cws_parse/readme.md rename to reproduction/joint_cws_parse/README.md diff --git a/reproduction/joint_cws_parse/models/CharParser.py b/reproduction/joint_cws_parse/models/CharParser.py index 1ed5ea2d..c07c070e 100644 --- a/reproduction/joint_cws_parse/models/CharParser.py +++ b/reproduction/joint_cws_parse/models/CharParser.py @@ -12,7 +12,7 @@ from torch.nn import functional as F from fastNLP.modules.dropout import TimestepDropout from fastNLP.modules.encoder.variational_rnn import VarLSTM from fastNLP import seq_len_to_mask -from fastNLP.modules import Embedding +from fastNLP.embeddings import Embedding def drop_input_independent(word_embeddings, dropout_emb): diff --git a/reproduction/joint_cws_parse/train.py b/reproduction/joint_cws_parse/train.py index 2f8b0d04..0c34614b 100644 --- a/reproduction/joint_cws_parse/train.py +++ b/reproduction/joint_cws_parse/train.py @@ -2,15 +2,15 @@ import sys sys.path.append('../..') from reproduction.joint_cws_parse.data.data_loader import CTBxJointLoader -from fastNLP.modules.encoder.embedding import StaticEmbedding +from fastNLP.embeddings.static_embedding import StaticEmbedding from torch import nn from functools import partial from reproduction.joint_cws_parse.models.CharParser import CharParser from reproduction.joint_cws_parse.models.metrics import SegAppCharParseF1Metric, CWSMetric -from fastNLP import cache_results, BucketSampler, Trainer +from fastNLP import BucketSampler, Trainer from torch import optim -from reproduction.joint_cws_parse.models.callbacks import DevCallback, OptimizerCallback -from torch.optim.lr_scheduler import LambdaLR, StepLR +from reproduction.joint_cws_parse.models.callbacks import DevCallback +from torch.optim.lr_scheduler import StepLR from fastNLP import Tester from fastNLP import GradientClipCallback, LRScheduler import os diff --git a/reproduction/Biaffine_parser/cfg.cfg b/reproduction/legacy/Biaffine_parser/cfg.cfg similarity index 100% rename from reproduction/Biaffine_parser/cfg.cfg rename to reproduction/legacy/Biaffine_parser/cfg.cfg diff --git a/reproduction/Biaffine_parser/infer.py b/reproduction/legacy/Biaffine_parser/infer.py similarity index 100% rename from reproduction/Biaffine_parser/infer.py rename to reproduction/legacy/Biaffine_parser/infer.py diff --git a/reproduction/Biaffine_parser/main.py b/reproduction/legacy/Biaffine_parser/main.py similarity index 100% rename from reproduction/Biaffine_parser/main.py rename to reproduction/legacy/Biaffine_parser/main.py diff --git a/reproduction/Biaffine_parser/run.py b/reproduction/legacy/Biaffine_parser/run.py similarity index 100% rename from reproduction/Biaffine_parser/run.py rename to reproduction/legacy/Biaffine_parser/run.py diff --git a/reproduction/Biaffine_parser/util.py b/reproduction/legacy/Biaffine_parser/util.py similarity index 100% rename from reproduction/Biaffine_parser/util.py rename to reproduction/legacy/Biaffine_parser/util.py diff --git a/reproduction/Chinese_word_segmentation/__init__.py b/reproduction/legacy/Chinese_word_segmentation/__init__.py similarity index 100% rename from reproduction/Chinese_word_segmentation/__init__.py rename to reproduction/legacy/Chinese_word_segmentation/__init__.py diff --git a/reproduction/Chinese_word_segmentation/cws.cfg b/reproduction/legacy/Chinese_word_segmentation/cws.cfg similarity index 100% rename from reproduction/Chinese_word_segmentation/cws.cfg rename to reproduction/legacy/Chinese_word_segmentation/cws.cfg diff --git a/reproduction/Chinese_word_segmentation/cws_io/__init__.py b/reproduction/legacy/Chinese_word_segmentation/cws_io/__init__.py similarity index 100% rename from reproduction/Chinese_word_segmentation/cws_io/__init__.py rename to reproduction/legacy/Chinese_word_segmentation/cws_io/__init__.py diff --git a/reproduction/Chinese_word_segmentation/cws_io/cws_reader.py b/reproduction/legacy/Chinese_word_segmentation/cws_io/cws_reader.py similarity index 100% rename from reproduction/Chinese_word_segmentation/cws_io/cws_reader.py rename to reproduction/legacy/Chinese_word_segmentation/cws_io/cws_reader.py diff --git a/reproduction/Chinese_word_segmentation/models/__init__.py b/reproduction/legacy/Chinese_word_segmentation/models/__init__.py similarity index 100% rename from reproduction/Chinese_word_segmentation/models/__init__.py rename to reproduction/legacy/Chinese_word_segmentation/models/__init__.py diff --git a/reproduction/Chinese_word_segmentation/models/cws_model.py b/reproduction/legacy/Chinese_word_segmentation/models/cws_model.py similarity index 98% rename from reproduction/Chinese_word_segmentation/models/cws_model.py rename to reproduction/legacy/Chinese_word_segmentation/models/cws_model.py index b41ad87d..0d10d2e5 100644 --- a/reproduction/Chinese_word_segmentation/models/cws_model.py +++ b/reproduction/legacy/Chinese_word_segmentation/models/cws_model.py @@ -4,7 +4,7 @@ from torch import nn from fastNLP.models.base_model import BaseModel from fastNLP.modules.decoder.mlp import MLP -from reproduction.Chinese_word_segmentation.utils import seq_lens_to_mask +from reproduction.legacy.Chinese_word_segmentation.utils import seq_lens_to_mask class CWSBiLSTMEncoder(BaseModel): diff --git a/reproduction/Chinese_word_segmentation/models/cws_transformer.py b/reproduction/legacy/Chinese_word_segmentation/models/cws_transformer.py similarity index 97% rename from reproduction/Chinese_word_segmentation/models/cws_transformer.py rename to reproduction/legacy/Chinese_word_segmentation/models/cws_transformer.py index e8ae5ecc..ae8a5a7f 100644 --- a/reproduction/Chinese_word_segmentation/models/cws_transformer.py +++ b/reproduction/legacy/Chinese_word_segmentation/models/cws_transformer.py @@ -9,7 +9,7 @@ from torch import nn import torch # from fastNLP.modules.encoder.transformer import TransformerEncoder -from reproduction.Chinese_word_segmentation.models.transformer import TransformerEncoder +from reproduction.legacy.Chinese_word_segmentation.models import TransformerEncoder from fastNLP.modules.decoder.crf import ConditionalRandomField,seq_len_to_byte_mask from fastNLP.modules.decoder.crf import allowed_transitions @@ -79,7 +79,7 @@ class TransformerCWS(nn.Module): return {'pred': probs, 'seq_lens':seq_lens} -from reproduction.Chinese_word_segmentation.models.dilated_transformer import TransformerDilateEncoder +from reproduction.legacy.Chinese_word_segmentation.models import TransformerDilateEncoder class TransformerDilatedCWS(nn.Module): def __init__(self, vocab_num, embed_dim=100, bigram_vocab_num=None, bigram_embed_dim=100, num_bigram_per_char=None, diff --git a/reproduction/Chinese_word_segmentation/process/__init__.py b/reproduction/legacy/Chinese_word_segmentation/process/__init__.py similarity index 100% rename from reproduction/Chinese_word_segmentation/process/__init__.py rename to reproduction/legacy/Chinese_word_segmentation/process/__init__.py diff --git a/reproduction/Chinese_word_segmentation/process/cws_processor.py b/reproduction/legacy/Chinese_word_segmentation/process/cws_processor.py similarity index 99% rename from reproduction/Chinese_word_segmentation/process/cws_processor.py rename to reproduction/legacy/Chinese_word_segmentation/process/cws_processor.py index 614d9ef5..1f64bed2 100644 --- a/reproduction/Chinese_word_segmentation/process/cws_processor.py +++ b/reproduction/legacy/Chinese_word_segmentation/process/cws_processor.py @@ -4,7 +4,7 @@ import re from fastNLP.api.processor import Processor from fastNLP.core.dataset import DataSet from fastNLP.core.vocabulary import Vocabulary -from reproduction.Chinese_word_segmentation.process.span_converter import SpanConverter +from reproduction.legacy.Chinese_word_segmentation.process.span_converter import SpanConverter _SPECIAL_TAG_PATTERN = '<[a-zA-Z]+>' diff --git a/reproduction/Chinese_word_segmentation/process/span_converter.py b/reproduction/legacy/Chinese_word_segmentation/process/span_converter.py similarity index 100% rename from reproduction/Chinese_word_segmentation/process/span_converter.py rename to reproduction/legacy/Chinese_word_segmentation/process/span_converter.py diff --git a/reproduction/Chinese_word_segmentation/utils.py b/reproduction/legacy/Chinese_word_segmentation/utils.py similarity index 100% rename from reproduction/Chinese_word_segmentation/utils.py rename to reproduction/legacy/Chinese_word_segmentation/utils.py diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/README.md b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/README.md similarity index 94% rename from reproduction/LSTM+self_attention_sentiment_analysis/README.md rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/README.md index 2dff7caa..dfb337ec 100644 --- a/reproduction/LSTM+self_attention_sentiment_analysis/README.md +++ b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/README.md @@ -1,5 +1,7 @@ # Prototype +这是一个很旧版本的reproduction，待修改 + ## Word2Idx.py A mapping model between words and indexes diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/Word2Idx.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/Word2Idx.py similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/Word2Idx.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/Word2Idx.py diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/config.cfg b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/config.cfg similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/config.cfg rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/config.cfg diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/dataloader.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/dataloader.py similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/dataloader.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/dataloader.py diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/example.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/example.py similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/example.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/example.py diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/main.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/main.py similarity index 90% rename from reproduction/LSTM+self_attention_sentiment_analysis/main.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/main.py index 871dc476..05077530 100644 --- a/reproduction/LSTM+self_attention_sentiment_analysis/main.py +++ b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/main.py @@ -1,6 +1,9 @@ +# 这是一个很旧版本的代码 + +""" import torch.nn.functional as F -from fastNLP.core.trainer import ClassificationTrainer +from fastNLP.core.trainer import Trainer from fastNLP.core.utils import ClassPreprocess as Preprocess from fastNLP.io.config_io import ConfigLoader from fastNLP.io.config_io import ConfigSection @@ -8,7 +11,7 @@ from fastNLP.io.dataset_loader import DummyClassificationReader as Dataset_loade from fastNLP.models.base_model import BaseModel from fastNLP.modules.aggregator.self_attention import SelfAttention from fastNLP.modules.decoder.mlp import MLP -from fastNLP.modules.encoder.embedding import Embedding as Embedding +from fastNLP.embeddings.embedding import Embedding as Embedding from fastNLP.modules.encoder.lstm import LSTM train_data_path = 'small_train_data.txt' @@ -61,12 +64,13 @@ class SELF_ATTENTION_YELP_CLASSIFICATION(BaseModel): train_args = ConfigSection() ConfigLoader("good path").load_config('config.cfg',{"train": train_args}) -train_args['vocab'] = len(word2index) +# train_args['vocab'] = len(word2index) -trainer = ClassificationTrainer(**train_args.data) +trainer = Trainer(**train_args.data) # for k in train_args.__dict__.keys(): # print(k, train_args[k]) model = SELF_ATTENTION_YELP_CLASSIFICATION(train_args) -trainer.train(model,train_data , dev_data) +trainer.train() +""" diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/predict.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/predict.py similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/predict.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/predict.py diff --git a/reproduction/LSTM+self_attention_sentiment_analysis/prepare.py b/reproduction/legacy/LSTM+self_attention_sentiment_analysis/prepare.py similarity index 100% rename from reproduction/LSTM+self_attention_sentiment_analysis/prepare.py rename to reproduction/legacy/LSTM+self_attention_sentiment_analysis/prepare.py diff --git a/reproduction/POS_tagging/pos_processor.py b/reproduction/legacy/POS_tagging/pos_processor.py similarity index 100% rename from reproduction/POS_tagging/pos_processor.py rename to reproduction/legacy/POS_tagging/pos_processor.py diff --git a/reproduction/POS_tagging/pos_reader.py b/reproduction/legacy/POS_tagging/pos_reader.py similarity index 100% rename from reproduction/POS_tagging/pos_reader.py rename to reproduction/legacy/POS_tagging/pos_reader.py diff --git a/reproduction/POS_tagging/pos_tag.cfg b/reproduction/legacy/POS_tagging/pos_tag.cfg similarity index 100% rename from reproduction/POS_tagging/pos_tag.cfg rename to reproduction/legacy/POS_tagging/pos_tag.cfg diff --git a/reproduction/POS_tagging/train_pos_tag.py b/reproduction/legacy/POS_tagging/train_pos_tag.py similarity index 100% rename from reproduction/POS_tagging/train_pos_tag.py rename to reproduction/legacy/POS_tagging/train_pos_tag.py diff --git a/reproduction/POS_tagging/utils.py b/reproduction/legacy/POS_tagging/utils.py similarity index 100% rename from reproduction/POS_tagging/utils.py rename to reproduction/legacy/POS_tagging/utils.py diff --git a/reproduction/matching/data/MatchingDataLoader.py b/reproduction/matching/data/MatchingDataLoader.py index 67fa4c8d..bba26a8a 100644 --- a/reproduction/matching/data/MatchingDataLoader.py +++ b/reproduction/matching/data/MatchingDataLoader.py @@ -1,3 +1,7 @@ +""" +这个文件的内容已合并到fastNLP.io.data_loader里，这个文件的内容不再更新 +""" + import os diff --git a/reproduction/matching/matching_bert.py b/reproduction/matching/matching_bert.py index 75112d5a..3ed75fd1 100644 --- a/reproduction/matching/matching_bert.py +++ b/reproduction/matching/matching_bert.py @@ -3,9 +3,8 @@ import numpy as np import torch from fastNLP.core import Trainer, Tester, AccuracyMetric, Const, Adam +from fastNLP.io.data_loader import SNLILoader, RTELoader, MNLILoader, QNLILoader, QuoraLoader -from reproduction.matching.data.MatchingDataLoader import SNLILoader, RTELoader, \ - MNLILoader, QNLILoader, QuoraLoader from reproduction.matching.model.bert import BertForNLI diff --git a/reproduction/matching/matching_cntn.py b/reproduction/matching/matching_cntn.py index d813164d..098f3bc4 100644 --- a/reproduction/matching/matching_cntn.py +++ b/reproduction/matching/matching_cntn.py @@ -1,11 +1,10 @@ import argparse import torch -import os from fastNLP.core import Trainer, Tester, Adam, AccuracyMetric, Const -from fastNLP.modules.encoder.embedding import StaticEmbedding +from fastNLP.embeddings import StaticEmbedding +from fastNLP.io.data_loader import QNLILoader, RTELoader, SNLILoader, MNLILoader -from reproduction.matching.data.MatchingDataLoader import QNLILoader, RTELoader, SNLILoader, MNLILoader from reproduction.matching.model.cntn import CNTNModel # define hyper-parameters diff --git a/reproduction/matching/matching_esim.py b/reproduction/matching/matching_esim.py index d878608f..2ff6916a 100644 --- a/reproduction/matching/matching_esim.py +++ b/reproduction/matching/matching_esim.py @@ -7,11 +7,10 @@ from torch.optim.lr_scheduler import StepLR from fastNLP.core import Trainer, Tester, AccuracyMetric, Const from fastNLP.core.callback import GradientClipCallback, LRScheduler -from fastNLP.modules.encoder.embedding import ElmoEmbedding, StaticEmbedding - -from reproduction.matching.data.MatchingDataLoader import SNLILoader, RTELoader, \ - MNLILoader, QNLILoader, QuoraLoader -from reproduction.matching.model.esim import ESIMModel +from fastNLP.embeddings.static_embedding import StaticEmbedding +from fastNLP.embeddings.elmo_embedding import ElmoEmbedding +from fastNLP.io.data_loader import SNLILoader, RTELoader, MNLILoader, QNLILoader, QuoraLoader +from fastNLP.models.snli import ESIM # define hyper-parameters @@ -81,7 +80,7 @@ else: raise RuntimeError(f'NOT support {arg.embedding} embedding yet!') # define model -model = ESIMModel(embedding, num_labels=len(data_info.vocabs[Const.TARGET])) +model = ESIM(embedding, num_labels=len(data_info.vocabs[Const.TARGET])) # define optimizer and callback optimizer = Adamax(lr=arg.lr, params=model.parameters()) diff --git a/reproduction/matching/matching_mwan.py b/reproduction/matching/matching_mwan.py index e96ee0c9..31af54c5 100644 --- a/reproduction/matching/matching_mwan.py +++ b/reproduction/matching/matching_mwan.py @@ -1,23 +1,17 @@ -import sys - -import os import random import numpy as np import torch -from torch.optim import Adadelta, SGD +from torch.optim import Adadelta from torch.optim.lr_scheduler import StepLR -from tqdm import tqdm - from fastNLP import CrossEntropyLoss from fastNLP import cache_results -from fastNLP.core import Trainer, Tester, Adam, AccuracyMetric, Const -from fastNLP.core.predictor import Predictor -from fastNLP.core.callback import GradientClipCallback, LRScheduler, FitlogCallback -from fastNLP.modules.encoder.embedding import ElmoEmbedding, StaticEmbedding +from fastNLP.core import Trainer, Tester, AccuracyMetric, Const +from fastNLP.core.callback import LRScheduler, FitlogCallback +from fastNLP.embeddings import StaticEmbedding -from fastNLP.io.data_loader import MNLILoader, QNLILoader, QuoraLoader, SNLILoader, RTELoader +from fastNLP.io.data_loader import MNLILoader, QNLILoader, SNLILoader, RTELoader from reproduction.matching.model.mwan import MwanModel import fitlog diff --git a/reproduction/matching/model/bert.py b/reproduction/matching/model/bert.py index 9b3a78b2..a21f8c36 100644 --- a/reproduction/matching/model/bert.py +++ b/reproduction/matching/model/bert.py @@ -4,7 +4,7 @@ import torch.nn as nn from fastNLP.core.const import Const from fastNLP.models import BaseModel -from fastNLP.modules.encoder.bert import BertModel +from fastNLP.embeddings.bert import BertModel class BertForNLI(BaseModel): diff --git a/reproduction/matching/model/cntn.py b/reproduction/matching/model/cntn.py index 0b4803fa..a0a104a3 100644 --- a/reproduction/matching/model/cntn.py +++ b/reproduction/matching/model/cntn.py @@ -6,7 +6,7 @@ import numpy as np from torch.nn import CrossEntropyLoss from fastNLP.models import BaseModel -from fastNLP.modules.encoder.embedding import TokenEmbedding +from fastNLP.embeddings.embedding import TokenEmbedding from fastNLP.core.const import Const diff --git a/reproduction/matching/model/esim.py b/reproduction/matching/model/esim.py index 187e565d..87e5ba65 100644 --- a/reproduction/matching/model/esim.py +++ b/reproduction/matching/model/esim.py @@ -5,8 +5,7 @@ import torch.nn.functional as F from torch.nn import CrossEntropyLoss from fastNLP.models import BaseModel -from fastNLP.modules.encoder.embedding import TokenEmbedding -from fastNLP.modules.encoder.lstm import LSTM +from fastNLP.embeddings.embedding import TokenEmbedding from fastNLP.core.const import Const from fastNLP.core.utils import seq_len_to_mask diff --git a/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py b/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py new file mode 100644 index 00000000..cec5ab76 --- /dev/null +++ b/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py @@ -0,0 +1,115 @@ + + +from fastNLP.io.base_loader import DataSetLoader, DataBundle +from fastNLP.io import ConllLoader +from reproduction.seqence_labelling.ner.data.utils import iob2bioes, iob2 +from fastNLP import Const +from reproduction.utils import check_dataloader_paths +from fastNLP import Vocabulary + +class ChineseNERLoader(DataSetLoader): + """ + 读取中文命名实体数据集，包括PeopleDaily, MSRA-NER, Weibo。数据在这里可以找到https://github.com/OYE93/Chinese-NLP-Corpus/tree/master/NER + 请确保输入数据的格式如下, 共两列，第一列为字，第二列为标签，不同句子以空行隔开 + 我 O + 们 O + 变 O + 而 O + 以 O + 书 O + 会 O + ... + + """ + def __init__(self, encoding_type:str='bioes'): + """ + + :param str encoding_type: 支持bio和bioes格式 + """ + super().__init__() + self._loader = ConllLoader(headers=['raw_chars', 'target'], indexes=[0, 1]) + + assert encoding_type in ('bio', 'bioes') + + self._tag_converters = [iob2] + if encoding_type == 'bioes': + self._tag_converters.append(iob2bioes) + + def load(self, path:str): + dataset = self._loader.load(path) + def convert_tag_schema(tags): + for converter in self._tag_converters: + tags = converter(tags) + return tags + if self._tag_converters: + dataset.apply_field(convert_tag_schema, field_name=Const.TARGET, new_field_name=Const.TARGET) + return dataset + + def process(self, paths, bigrams=False, trigrams=False): + """ + + :param paths: + :param bool, bigrams: 是否包含生成bigram feature, [a, b, c, d] -> [ab, bc, cd, d] + :param bool, trigrams: 是否包含trigram feature，[a, b, c, d] -> [abc, bcd, cd, d] + :return: DataBundle + 包含以下的fields + raw_chars: List[str] + chars: List[int] + seq_len: int, 字的长度 + bigrams: List[int], optional + trigrams: List[int], optional + target: List[int] + """ + paths = check_dataloader_paths(paths) + data = DataBundle() + input_fields = [Const.CHAR_INPUT, Const.INPUT_LEN, Const.TARGET] + target_fields = [Const.TARGET, Const.INPUT_LEN] + + for name, path in paths.items(): + dataset = self.load(path) + if bigrams: + dataset.apply_field(lambda raw_chars: [c1+c2 for c1, c2 in zip(raw_chars, raw_chars[1:]+[''])], + field_name='raw_chars', new_field_name='bigrams') + + if trigrams: + dataset.apply_field(lambda raw_chars: [c1+c2+c3 for c1, c2, c3 in zip(raw_chars, + raw_chars[1:]+[''], + raw_chars[2:]+['']*2)], + field_name='raw_chars', new_field_name='trigrams') + data.datasets[name] = dataset + + char_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='raw_chars', + no_create_entry_dataset=[dataset for name, dataset in data.datasets.items() if name!='train']) + char_vocab.index_dataset(*data.datasets.values(), field_name='raw_chars', new_field_name=Const.CHAR_INPUT) + data.vocabs[Const.CHAR_INPUT] = char_vocab + + target_vocab = Vocabulary(unknown=None, padding=None).from_dataset(data.datasets['train'], field_name=Const.TARGET) + target_vocab.index_dataset(*data.datasets.values(), field_name=Const.TARGET) + data.vocabs[Const.TARGET] = target_vocab + + if bigrams: + bigram_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='bigrams', + no_create_entry_dataset=[dataset for name, dataset in + data.datasets.items() if name != 'train']) + bigram_vocab.index_dataset(*data.datasets.values(), field_name='bigrams', new_field_name='bigrams') + data.vocabs['bigrams'] = bigram_vocab + input_fields.append('bigrams') + + if trigrams: + trigram_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='trigrams', + no_create_entry_dataset=[dataset for name, dataset in + data.datasets.items() if name != 'train']) + trigram_vocab.index_dataset(*data.datasets.values(), field_name='trigrams', new_field_name='trigrams') + data.vocabs['trigrams'] = trigram_vocab + input_fields.append('trigrams') + + for name, dataset in data.datasets.items(): + dataset.add_seq_len(Const.CHAR_INPUT) + dataset.set_input(*input_fields) + dataset.set_target(*target_fields) + + return data + + + + diff --git a/reproduction/seqence_labelling/chinese_ner/data/__init__.py b/reproduction/seqence_labelling/chinese_ner/data/__init__.py new file mode 100644 index 00000000..e69de29b diff --git a/reproduction/seqence_labelling/chinese_ner/train_bert.py b/reproduction/seqence_labelling/chinese_ner/train_bert.py new file mode 100644 index 00000000..a34b7d01 --- /dev/null +++ b/reproduction/seqence_labelling/chinese_ner/train_bert.py @@ -0,0 +1,78 @@ + + +""" +使用Bert进行中文命名实体识别 + +""" + +import sys + +sys.path.append('../../../') + +from torch import nn + +from fastNLP.embeddings import BertEmbedding, Embedding +from reproduction.seqence_labelling.chinese_ner.data.ChineseNER import ChineseNERLoader +from fastNLP import Trainer, Const +from fastNLP import BucketSampler, SpanFPreRecMetric, GradientClipCallback +from fastNLP.modules import MLP +from fastNLP.core.callback import WarmupCallback +from fastNLP import CrossEntropyLoss +from fastNLP.core.optimizer import AdamW +import os + +from fastNLP import cache_results + +encoding_type = 'bio' + +@cache_results('caches/msra.pkl') +def get_data(): + data = ChineseNERLoader(encoding_type=encoding_type).process("MSRA/") + return data +data = get_data() +print(data) + +class BertCNNER(nn.Module): + def __init__(self, embed, tag_size): + super().__init__() + + self.embedding = Embedding(embed, dropout=0.1) + self.tag_size = tag_size + self.mlp = MLP(size_layer=[self.embedding.embedding_dim, tag_size]) + def forward(self, chars): + # batch_size, max_len = words.size() + chars = self.embedding(chars) + outputs = self.mlp(chars) + + return {Const.OUTPUT: outputs} + +embed = BertEmbedding(data.vocabs[Const.CHAR_INPUT], model_dir_or_name='en-base', + pool_method='max', requires_grad=True, layers='11') + +for name, dataset in data.datasets.items(): + dataset.set_pad_val(Const.TARGET, -100) + +callbacks = [ + GradientClipCallback(clip_type='norm', clip_value=1), + WarmupCallback(warmup=0.1, schedule='linear') + ] + +model = BertCNNER(embed, len(data.vocabs[Const.TARGET])) +optimizer = AdamW(model.parameters(), lr=1e-4) + +for name, dataset in data.datasets.items(): + original_len = len(dataset) + dataset.drop(lambda x:x['seq_len']>256, inplace=True) + clipped_len = len(dataset) + print("Delete {} instances in {}.".format(original_len-clipped_len, name)) + +os.environ['CUDA_VISIBLE_DEVICES'] = '0,1' + +trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=optimizer, sampler=BucketSampler(), + device=[0, 1], dev_data=data.datasets['test'], batch_size=20, + metrics=SpanFPreRecMetric(tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type), + loss=CrossEntropyLoss(reduction='sum'), + callbacks=callbacks, num_workers=2, n_epochs=5, + check_code_level=-1, update_every=3) +trainer.train() + diff --git a/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py b/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py new file mode 100644 index 00000000..53a85186 --- /dev/null +++ b/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py @@ -0,0 +1,94 @@ + + + +from reproduction.seqence_labelling.chinese_ner.data.ChineseNER import ChineseNERLoader +from fastNLP.embeddings import StaticEmbedding + +from torch import nn +import torch +from fastNLP.embeddings.utils import get_embeddings +from fastNLP.modules import LSTM +from fastNLP.modules import ConditionalRandomField +from fastNLP.modules import allowed_transitions +import torch.nn.functional as F +from fastNLP import seq_len_to_mask +from fastNLP.core.const import Const as C +from fastNLP import SpanFPreRecMetric, Trainer +from fastNLP import cache_results + +class CNBiLSTMCRFNER(nn.Module): + def __init__(self, char_embed, num_classes, bigram_embed=None, trigram_embed=None, num_layers=1, hidden_size=100, + dropout=0.5, target_vocab=None, encoding_type=None): + super().__init__() + + self.char_embed = get_embeddings(char_embed) + embed_size = self.char_embed.embedding_dim + if bigram_embed: + self.bigram_embed = get_embeddings(bigram_embed) + embed_size += self.bigram_embed.embedding_dim + if trigram_embed: + self.trigram_ebmbed = get_embeddings(trigram_embed) + embed_size += self.bigram_embed.embedding_dim + + if num_layers>1: + self.lstm = LSTM(embed_size, num_layers=num_layers, hidden_size=hidden_size//2, bidirectional=True, + batch_first=True, dropout=dropout) + else: + self.lstm = LSTM(embed_size, num_layers=num_layers, hidden_size=hidden_size//2, bidirectional=True, + batch_first=True) + + self.dropout = nn.Dropout(dropout) + self.fc = nn.Linear(hidden_size, num_classes) + + trans = None + if target_vocab is not None and encoding_type is not None: + trans = allowed_transitions(target_vocab.idx2word, encoding_type=encoding_type, include_start_end=True) + + self.crf = ConditionalRandomField(num_classes, include_start_end_trans=True, allowed_transitions=trans) + + def _forward(self, chars, bigrams=None, trigrams=None, seq_len=None, target=None): + chars = self.char_embed(chars) + if hasattr(self, 'bigram_embed'): + bigrams = self.bigram_embed(bigrams) + chars = torch.cat((chars, bigrams), dim=-1) + if hasattr(self, 'trigram_embed'): + trigrams = self.trigram_embed(trigrams) + chars = torch.cat((chars, trigrams), dim=-1) + feats, _ = self.lstm(chars, seq_len=seq_len) + feats = self.fc(feats) + feats = self.dropout(feats) + logits = F.log_softmax(feats, dim=-1) + mask = seq_len_to_mask(seq_len) + if target is None: + pred, _ = self.crf.viterbi_decode(logits, mask) + return {C.OUTPUT: pred} + else: + loss = self.crf(logits, target, mask).mean() + return {C.LOSS:loss} + + def forward(self, chars, target, bigrams=None, trigrams=None, seq_len=None): + return self._forward(chars, bigrams, trigrams, seq_len, target) + + def predict(self, chars, seq_len=None, bigrams=None, trigrams=None): + return self._forward(chars, bigrams, trigrams, seq_len) + +# data_bundle = pickle.load(open('caches/msra.pkl', 'rb')) +@cache_results('caches/msra.pkl', _refresh=True) +def get_data(): + data_bundle = ChineseNERLoader().process('MSRA-NER/', bigrams=True) + char_embed = StaticEmbedding(data_bundle.vocabs['chars'], + model_dir_or_name='cn-char') + bigram_embed = StaticEmbedding(data_bundle.vocabs['bigrams'], + model_dir_or_name='cn-bigram') + return data_bundle, char_embed, bigram_embed +data_bundle, char_embed, bigram_embed = get_data() +print(data_bundle) +# exit(0) +data_bundle.datasets['train'].set_input('target') +data_bundle.datasets['dev'].set_input('target') +model = CNBiLSTMCRFNER(char_embed, num_classes=len(data_bundle.vocabs['target']), bigram_embed=bigram_embed) + +Trainer(data_bundle.datasets['train'], model, batch_size=640, + metrics=SpanFPreRecMetric(data_bundle.vocabs['target'], encoding_type='bioes'), + num_workers=2, dev_data=data_bundle. datasets['dev'], device=3).train() + diff --git a/reproduction/seqence_labelling/cws/model/model.py b/reproduction/seqence_labelling/cws/model/model.py index bdd9002d..de945ac3 100644 --- a/reproduction/seqence_labelling/cws/model/model.py +++ b/reproduction/seqence_labelling/cws/model/model.py @@ -1,6 +1,6 @@ from torch import nn import torch -from fastNLP.modules import Embedding +from fastNLP.embeddings import Embedding import numpy as np from reproduction.seqence_labelling.cws.model.module import FeatureFunMax, SemiCRFShiftRelay from fastNLP.modules import LSTM diff --git a/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py b/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py index f1ff83d8..a6070f39 100644 --- a/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py +++ b/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py @@ -6,7 +6,7 @@ from fastNLP import Vocabulary from fastNLP import Const from reproduction.utils import check_dataloader_paths -from fastNLP.io.dataset_loader import ConllLoader +from fastNLP.io import ConllLoader from reproduction.seqence_labelling.ner.data.utils import iob2bioes, iob2 class OntoNoteNERDataLoader(DataSetLoader): diff --git a/reproduction/seqence_labelling/ner/model/dilated_cnn.py b/reproduction/seqence_labelling/ner/model/dilated_cnn.py index bf661354..89d51d56 100644 --- a/reproduction/seqence_labelling/ner/model/dilated_cnn.py +++ b/reproduction/seqence_labelling/ner/model/dilated_cnn.py @@ -106,7 +106,9 @@ class IDCNN(nn.Module): if self.crf is not None and target is not None: loss = self.crf(y.transpose(1, 2), t, mask) else: - # t.masked_fill_(mask == 0, -100) + y.masked_fill_((mask == 0)[:,None,:], -100) + # f_mask = mask.float() + # t = f_mask * t + (1-f_mask) * -100 loss = F.cross_entropy(y, t, ignore_index=-100) return loss diff --git a/reproduction/seqence_labelling/ner/train_cnn_lstm_crf_conll2003.py b/reproduction/seqence_labelling/ner/train_cnn_lstm_crf_conll2003.py index e9d18048..caa0247a 100644 --- a/reproduction/seqence_labelling/ner/train_cnn_lstm_crf_conll2003.py +++ b/reproduction/seqence_labelling/ner/train_cnn_lstm_crf_conll2003.py @@ -1,7 +1,7 @@ import sys sys.path.append('../../..') -from fastNLP.modules.encoder.embedding import CNNCharEmbedding, StaticEmbedding, BertEmbedding, ElmoEmbedding, StackEmbedding +from fastNLP.embeddings.embedding import CNNCharEmbedding, StaticEmbedding from fastNLP.core.vocabulary import VocabularyOption from reproduction.seqence_labelling.ner.model.lstm_cnn_crf import CNNBiLSTMCRF @@ -9,13 +9,11 @@ from fastNLP import Trainer from fastNLP import SpanFPreRecMetric from fastNLP import BucketSampler from fastNLP import Const -from torch.optim import SGD, Adam +from torch.optim import SGD from fastNLP import GradientClipCallback from fastNLP.core.callback import FitlogCallback, LRScheduler from torch.optim.lr_scheduler import LambdaLR -from fastNLP.core.optimizer import AdamW # from reproduction.seqence_labelling.ner.model.swats import SWATS -from reproduction.seqence_labelling.chinese_ner.callbacks import SaveModelCallback from fastNLP import cache_results import fitlog diff --git a/reproduction/seqence_labelling/ner/train_idcnn.py b/reproduction/seqence_labelling/ner/train_idcnn.py index 7de8a61c..53f2798f 100644 --- a/reproduction/seqence_labelling/ner/train_idcnn.py +++ b/reproduction/seqence_labelling/ner/train_idcnn.py @@ -1,20 +1,18 @@ from reproduction.seqence_labelling.ner.data.OntoNoteLoader import OntoNoteNERDataLoader -from reproduction.seqence_labelling.ner.data.Conll2003Loader import Conll2003DataLoader -from fastNLP.core.callback import FitlogCallback, LRScheduler +from fastNLP.core.callback import LRScheduler from fastNLP import GradientClipCallback -from torch.optim.lr_scheduler import LambdaLR, CosineAnnealingLR -from torch.optim import SGD, Adam +from torch.optim.lr_scheduler import LambdaLR +from torch.optim import Adam from fastNLP import Const -from fastNLP import RandomSampler, BucketSampler +from fastNLP import BucketSampler from fastNLP import SpanFPreRecMetric -from fastNLP import Trainer +from fastNLP import Trainer, Tester from fastNLP.core.metrics import MetricBase from reproduction.seqence_labelling.ner.model.dilated_cnn import IDCNN from fastNLP.core.utils import Option -from fastNLP.modules.encoder.embedding import CNNCharEmbedding, StaticEmbedding +from fastNLP.embeddings.embedding import StaticEmbedding from fastNLP.core.utils import cache_results from fastNLP.core.vocabulary import VocabularyOption -import sys import torch.cuda import os os.environ['FASTNLP_BASE_URL'] = 'http://10.141.222.118:8888/file/download/' @@ -31,7 +29,7 @@ def get_path(path): ops = Option( batch_size=128, num_epochs=100, - lr=5e-4, + lr=3e-4, repeats=3, num_layers=3, num_filters=400, @@ -39,18 +37,18 @@ ops = Option( gradient_clip=5, ) -@cache_results('ontonotes-min_freq0-case-cache') +@cache_results('ontonotes-case-cache') def load_data(): print('loading data') - # data = OntoNoteNERDataLoader(encoding_type=encoding_type).process( - # data_path = get_path('workdir/datasets/ontonotes-v4') - # lower=False, - # word_vocab_opt=VocabularyOption(min_freq=0), - # ) - data = Conll2003DataLoader(task='ner', encoding_type=encoding_type).process( - paths=get_path('workdir/datasets/conll03'), - lower=False, word_vocab_opt=VocabularyOption(min_freq=0) + data = OntoNoteNERDataLoader(encoding_type=encoding_type).process( + paths = get_path('workdir/datasets/ontonotes-v4'), + lower=False, + word_vocab_opt=VocabularyOption(min_freq=0), ) + # data = Conll2003DataLoader(task='ner', encoding_type=encoding_type).process( + # paths=get_path('workdir/datasets/conll03'), + # lower=False, word_vocab_opt=VocabularyOption(min_freq=0) + # ) # char_embed = CNNCharEmbedding(vocab=data.vocabs['cap_words'], embed_size=30, char_emb_size=30, filter_nums=[30], # kernel_sizes=[3]) @@ -88,11 +86,11 @@ model = IDCNN(init_embed=word_embed, kernel_size=3, use_crf=ops.use_crf, use_projection=True, block_loss=True, - input_dropout=0.5, hidden_dropout=0.0, inner_dropout=0.0) + input_dropout=0.5, hidden_dropout=0.2, inner_dropout=0.2) print(model) -callbacks = [GradientClipCallback(clip_value=ops.gradient_clip, clip_type='norm'),] +callbacks = [GradientClipCallback(clip_value=ops.gradient_clip, clip_type='value'),] metrics = [] metrics.append( SpanFPreRecMetric( @@ -123,8 +121,9 @@ metrics.append( LossMetric(loss=Const.LOSS) ) -optimizer = Adam(model.parameters(), lr=ops.lr, weight_decay=1e-4) -# scheduler = LRScheduler(LambdaLR(optimizer, lr_lambda=lambda epoch: 1 / (1 + 0.05 * epoch))) +optimizer = Adam(model.parameters(), lr=ops.lr, weight_decay=0) +scheduler = LRScheduler(LambdaLR(optimizer, lr_lambda=lambda epoch: 1 / (1 + 0.05 * epoch))) +callbacks.append(scheduler) # callbacks.append(LRScheduler(CosineAnnealingLR(optimizer, 15))) # optimizer = SWATS(model.parameters(), verbose=True) # optimizer = Adam(model.parameters(), lr=0.005) @@ -138,3 +137,16 @@ trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=opti check_code_level=-1, callbacks=callbacks, num_workers=2, n_epochs=ops.num_epochs) trainer.train() + +torch.save(model, 'idcnn.pt') + +tester = Tester( + data=data.datasets['test'], + model=model, + metrics=metrics, + batch_size=ops.batch_size, + num_workers=2, + device=device +) +tester.test() + diff --git a/reproduction/seqence_labelling/ner/train_ontonote.py b/reproduction/seqence_labelling/ner/train_ontonote.py index 6548cb9f..894d42ce 100644 --- a/reproduction/seqence_labelling/ner/train_ontonote.py +++ b/reproduction/seqence_labelling/ner/train_ontonote.py @@ -2,64 +2,81 @@ import sys sys.path.append('../../..') -from fastNLP.modules.encoder.embedding import CNNCharEmbedding, StaticEmbedding +from fastNLP.embeddings import CNNCharEmbedding, StaticEmbedding from reproduction.seqence_labelling.ner.model.lstm_cnn_crf import CNNBiLSTMCRF from fastNLP import Trainer from fastNLP import SpanFPreRecMetric -from fastNLP import BucketSampler from fastNLP import Const -from torch.optim import SGD, Adam +from torch.optim import SGD from torch.optim.lr_scheduler import LambdaLR from fastNLP import GradientClipCallback +from fastNLP.core.vocabulary import VocabularyOption from fastNLP.core.callback import FitlogCallback, LRScheduler -from reproduction.seqence_labelling.ner.model.swats import SWATS +from functools import partial +from torch import nn +from fastNLP import cache_results import fitlog fitlog.debug() +fitlog.set_log_dir('logs/') + +fitlog.add_hyper_in_file(__file__) +#######hyper +normalize = False +divide_std = True +lower = False +lr = 0.015 +dropout = 0.5 +batch_size = 20 +init_method = 'default' +job_embed = False +data_name = 'ontonote' +#######hyper + + +init_method = {'default': None, + 'xavier': partial(nn.init.xavier_normal_, gain=0.02), + 'normal': partial(nn.init.normal_, std=0.02) + }[init_method] + from reproduction.seqence_labelling.ner.data.OntoNoteLoader import OntoNoteNERDataLoader encoding_type = 'bioes' -data = OntoNoteNERDataLoader(encoding_type=encoding_type).process('/hdd/fudanNLP/fastNLP/others/data/v4/english', - lower=True) - -import joblib -raw_data = joblib.load('/hdd/fudanNLP/fastNLP/others/NER-with-LS/data/ontonotes_with_data.joblib') -def convert_to_ids(raw_words): - ids = [] - for word in raw_words: - id = raw_data['word_to_id'][word] - id = raw_data['id_to_emb_map'][id] - ids.append(id) - return ids -word_embed = raw_data['emb_matrix'] -for name, dataset in data.datasets.items(): - dataset.apply_field(convert_to_ids, field_name='raw_words', new_field_name=Const.INPUT) +@cache_results('caches/ontonotes.pkl') +def cache(): + data = OntoNoteNERDataLoader(encoding_type=encoding_type).process('../../../../others/data/v4/english', + lower=lower, + word_vocab_opt=VocabularyOption(min_freq=1)) + char_embed = CNNCharEmbedding(vocab=data.vocabs['cap_words'], embed_size=30, char_emb_size=30, filter_nums=[30], + kernel_sizes=[3]) + word_embed = StaticEmbedding(vocab=data.vocabs[Const.INPUT], + model_dir_or_name='/remote-home/hyan01/fastnlp_caches/glove.6B.100d/glove.6B.100d.txt', + requires_grad=True, + normalize=normalize, + init_method=init_method) + return data, char_embed, word_embed +data, char_embed, word_embed = cache() print(data) -char_embed = CNNCharEmbedding(vocab=data.vocabs['cap_words'], embed_size=30, char_emb_size=30, filter_nums=[30], - kernel_sizes=[3]) -# word_embed = StaticEmbedding(vocab=data.vocabs[Const.INPUT], -# model_dir_or_name='/hdd/fudanNLP/pretrain_vectors/glove.6B.100d.txt', -# requires_grad=True) model = CNNBiLSTMCRF(word_embed, char_embed, hidden_size=1200, num_layers=1, tag_vocab=data.vocabs[Const.TARGET], - encoding_type=encoding_type) + encoding_type=encoding_type, dropout=dropout) -callbacks = [GradientClipCallback(clip_value=5, clip_type='value'), - FitlogCallback(data.datasets['test'], verbose=1)] +callbacks = [ + GradientClipCallback(clip_value=5, clip_type='value'), + FitlogCallback(data.datasets['test'], verbose=1) + ] -optimizer = SGD(model.parameters(), lr=0.01, momentum=0.9) +optimizer = SGD(model.parameters(), lr=lr, momentum=0.9) scheduler = LRScheduler(LambdaLR(optimizer, lr_lambda=lambda epoch: 1 / (1 + 0.05 * epoch))) callbacks.append(scheduler) -# optimizer = SWATS(model.parameters(), verbose=True) -# optimizer = Adam(model.parameters(), lr=0.005) -trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=optimizer, sampler=BucketSampler(num_buckets=100), - device=0, dev_data=data.datasets['dev'], batch_size=10, +trainer = Trainer(train_data=data.datasets['dev'][:100], model=model, optimizer=optimizer, sampler=None, + device=0, dev_data=data.datasets['dev'][:100], batch_size=batch_size, metrics=SpanFPreRecMetric(tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type), callbacks=callbacks, num_workers=1, n_epochs=100) trainer.train() \ No newline at end of file diff --git a/reproduction/text_classification/data/IMDBLoader.py b/reproduction/text_classification/data/IMDBLoader.py index d57ee41b..94244431 100644 --- a/reproduction/text_classification/data/IMDBLoader.py +++ b/reproduction/text_classification/data/IMDBLoader.py @@ -10,7 +10,6 @@ from fastNLP import Const from functools import partial from reproduction.utils import check_dataloader_paths, get_tokenizer - class IMDBLoader(DataSetLoader): """ 读取IMDB数据集，DataSet包含以下fields: @@ -51,6 +50,7 @@ class IMDBLoader(DataSetLoader): datasets = {} info = DataBundle() + paths = check_dataloader_paths(paths) for name, path in paths.items(): dataset = self.load(path) datasets[name] = dataset diff --git a/reproduction/text_classification/model/HAN.py b/reproduction/text_classification/model/HAN.py index 0902d1e4..7ebbe30f 100644 --- a/reproduction/text_classification/model/HAN.py +++ b/reproduction/text_classification/model/HAN.py @@ -1,7 +1,7 @@ import torch import torch.nn as nn from torch.autograd import Variable -from fastNLP.modules.utils import get_embeddings +from fastNLP.embeddings.utils import get_embeddings from fastNLP.core import Const as C diff --git a/reproduction/text_classification/model/awd_lstm.py b/reproduction/text_classification/model/awd_lstm.py index 0d8f711a..c9c8a153 100644 --- a/reproduction/text_classification/model/awd_lstm.py +++ b/reproduction/text_classification/model/awd_lstm.py @@ -2,7 +2,7 @@ import torch import torch.nn as nn from fastNLP.core.const import Const as C from .awdlstm_module import LSTM -from fastNLP.modules import encoder +from fastNLP.embeddings.utils import get_embeddings from fastNLP.modules.decoder.mlp import MLP @@ -14,7 +14,7 @@ class AWDLSTMSentiment(nn.Module): nfc=128, wdrop=0.5): super(AWDLSTMSentiment,self).__init__() - self.embed = encoder.Embedding(init_embed) + self.embed = get_embeddings(init_embed) self.lstm = LSTM(input_size=self.embed.embedding_dim, hidden_size=hidden_dim, num_layers=num_layers, bidirectional=True, wdrop=wdrop) self.mlp = MLP(size_layer=[hidden_dim* 2, nfc, num_classes]) diff --git a/reproduction/text_classification/model/dpcnn.py b/reproduction/text_classification/model/dpcnn.py index dafe62bc..ae2d46bd 100644 --- a/reproduction/text_classification/model/dpcnn.py +++ b/reproduction/text_classification/model/dpcnn.py @@ -1,6 +1,6 @@ import torch import torch.nn as nn -from fastNLP.modules.utils import get_embeddings +from fastNLP.embeddings.utils import get_embeddings from fastNLP.core import Const as C diff --git a/reproduction/text_classification/model/lstm.py b/reproduction/text_classification/model/lstm.py index 388f3f1c..16c7652c 100644 --- a/reproduction/text_classification/model/lstm.py +++ b/reproduction/text_classification/model/lstm.py @@ -2,7 +2,7 @@ import torch import torch.nn as nn from fastNLP.core.const import Const as C from fastNLP.modules.encoder.lstm import LSTM -from fastNLP.modules import encoder +from fastNLP.embeddings.utils import get_embeddings from fastNLP.modules.decoder.mlp import MLP @@ -13,14 +13,14 @@ class BiLSTMSentiment(nn.Module): num_layers=1, nfc=128): super(BiLSTMSentiment,self).__init__() - self.embed = encoder.Embedding(init_embed) + self.embed = get_embeddings(init_embed) self.lstm = LSTM(input_size=self.embed.embedding_dim, hidden_size=hidden_dim, num_layers=num_layers, bidirectional=True) - self.mlp = MLP(size_layer=[hidden_dim* 2, nfc, num_classes]) + self.mlp = MLP(size_layer=[hidden_dim*2, nfc, num_classes]) def forward(self, words): x_emb = self.embed(words) output, _ = self.lstm(x_emb) - output = self.mlp(output[:,-1,:]) + output = self.mlp(torch.max(output, dim=1)[0]) return {C.OUTPUT: output} def predict(self, words): diff --git a/reproduction/text_classification/model/lstm_self_attention.py b/reproduction/text_classification/model/lstm_self_attention.py index 239635fe..9a39049d 100644 --- a/reproduction/text_classification/model/lstm_self_attention.py +++ b/reproduction/text_classification/model/lstm_self_attention.py @@ -2,8 +2,8 @@ import torch import torch.nn as nn from fastNLP.core.const import Const as C from fastNLP.modules.encoder.lstm import LSTM -from fastNLP.modules import encoder -from fastNLP.modules.aggregator.attention import SelfAttention +from fastNLP.embeddings.utils import get_embeddings +from fastNLP.modules.encoder.attention import SelfAttention from fastNLP.modules.decoder.mlp import MLP @@ -16,7 +16,7 @@ class BiLSTM_SELF_ATTENTION(nn.Module): attention_hops=1, nfc=128): super(BiLSTM_SELF_ATTENTION,self).__init__() - self.embed = encoder.Embedding(init_embed) + self.embed = get_embeddings(init_embed) self.lstm = LSTM(input_size=self.embed.embedding_dim, hidden_size=hidden_dim, num_layers=num_layers, bidirectional=True) self.attention = SelfAttention(input_size=hidden_dim * 2 , attention_unit=attention_unit, attention_hops=attention_hops) self.mlp = MLP(size_layer=[hidden_dim* 2*attention_hops, nfc, num_classes]) diff --git a/reproduction/text_classification/train_HAN.py b/reproduction/text_classification/train_HAN.py index b1135342..a8b06146 100644 --- a/reproduction/text_classification/train_HAN.py +++ b/reproduction/text_classification/train_HAN.py @@ -9,11 +9,9 @@ os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" from fastNLP.core.const import Const as C from fastNLP.core import LRScheduler -import torch.nn as nn -from fastNLP.io.dataset_loader import SSTLoader -from reproduction.text_classification.data.yelpLoader import yelpLoader +from fastNLP.io.data_loader import YelpLoader from reproduction.text_classification.model.HAN import HANCLS -from fastNLP.modules.encoder.embedding import StaticEmbedding, CNNCharEmbedding, StackEmbedding +from fastNLP.embeddings import StaticEmbedding from fastNLP import CrossEntropyLoss, AccuracyMetric from fastNLP.core.trainer import Trainer from torch.optim import SGD @@ -44,7 +42,7 @@ ops = Config() ##1.task相关信息：利用dataloader载入dataInfo -datainfo = yelpLoader(fine_grained=True).process(paths=ops.datapath, train_ds=['train']) +datainfo = YelpLoader(fine_grained=True).process(paths=ops.datapath, train_ds=['train']) print(len(datainfo.datasets['train'])) print(len(datainfo.datasets['test'])) diff --git a/reproduction/text_classification/train_awdlstm.py b/reproduction/text_classification/train_awdlstm.py index 007b2910..b2a67fdb 100644 --- a/reproduction/text_classification/train_awdlstm.py +++ b/reproduction/text_classification/train_awdlstm.py @@ -5,20 +5,13 @@ import os os.environ['FASTNLP_BASE_URL'] = 'http://10.141.222.118:8888/file/download/' os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' - -import torch.nn as nn - -from data.IMDBLoader import IMDBLoader -from fastNLP.modules.encoder.embedding import StaticEmbedding +from fastNLP.io.data_loader import IMDBLoader +from fastNLP.embeddings import StaticEmbedding from model.awd_lstm import AWDLSTMSentiment -from fastNLP.core.const import Const as C from fastNLP import CrossEntropyLoss, AccuracyMetric -from fastNLP import Trainer, Tester +from fastNLP import Trainer from torch.optim import Adam -from fastNLP.io.model_io import ModelLoader, ModelSaver - -import argparse class Config(): diff --git a/reproduction/text_classification/train_bert.py b/reproduction/text_classification/train_bert.py index e69de29b..25337d9e 100644 --- a/reproduction/text_classification/train_bert.py +++ b/reproduction/text_classification/train_bert.py @@ -0,0 +1,33 @@ +import sys +sys.path.append('../../') + +from reproduction.text_classification.data.IMDBLoader import IMDBLoader +from fastNLP.embeddings import BertEmbedding +from reproduction.text_classification.model.lstm import BiLSTMSentiment +from fastNLP import Trainer +from fastNLP import CrossEntropyLoss, AccuracyMetric +from fastNLP import cache_results +from fastNLP import Tester + +# 对返回结果进行缓存，下一次运行就会自动跳过预处理 +@cache_results('imdb.pkl') +def get_data(): + data_bundle = IMDBLoader().process('imdb/') + return data_bundle +data_bundle = get_data() + +print(data_bundle) + +# 删除超过512, 但由于英语中会把word进行word piece处理，所以截取的时候做一点的裕量 +data_bundle.datasets['train'].drop(lambda x:len(x['words'])>400) +data_bundle.datasets['dev'].drop(lambda x:len(x['words'])>400) +data_bundle.datasets['test'].drop(lambda x:len(x['words'])>400) +bert_embed = BertEmbedding(data_bundle.vocabs['words'], requires_grad=False, + model_dir_or_name="en-base-uncased") +model = BiLSTMSentiment(bert_embed, len(data_bundle.vocabs['target'])) + +Trainer(data_bundle.datasets['train'], model, optimizer=None, loss=CrossEntropyLoss(), device=0, + batch_size=10, dev_data=data_bundle.datasets['dev'], metrics=AccuracyMetric()).train() + +# 在测试集上测试一下效果 +Tester(data_bundle.datasets['test'], model, batch_size=32, metrics=AccuracyMetric()).test() \ No newline at end of file diff --git a/reproduction/text_classification/train_char_cnn.py b/reproduction/text_classification/train_char_cnn.py index e4bb9220..0b8fc535 100644 --- a/reproduction/text_classification/train_char_cnn.py +++ b/reproduction/text_classification/train_char_cnn.py @@ -7,23 +7,17 @@ import sys sys.path.append('../..') from fastNLP.core.const import Const as C import torch.nn as nn -from data.yelpLoader import yelpLoader +from fastNLP.io.data_loader import YelpLoader #from data.sstLoader import sst2Loader -from fastNLP.io.data_loader.sst import SST2Loader -from data.IMDBLoader import IMDBLoader from model.char_cnn import CharacterLevelCNN -from fastNLP.core.vocabulary import Vocabulary -from fastNLP.models.cnn_text_classification import CNNText -from fastNLP.modules.encoder.embedding import CNNCharEmbedding,StaticEmbedding,StackEmbedding,LSTMCharEmbedding from fastNLP import CrossEntropyLoss, AccuracyMetric from fastNLP.core.trainer import Trainer from torch.optim import SGD from torch.autograd import Variable import torch -from fastNLP import BucketSampler -from torch.optim.lr_scheduler import CosineAnnealingLR, LambdaLR +from torch.optim.lr_scheduler import LambdaLR from fastNLP.core import LRScheduler -from utils.util_init import set_rng_seeds + ##hyper #todo 这里加入fastnlp的记录 @@ -117,7 +111,7 @@ ops=Config ##1.task相关信息：利用dataloader载入dataInfo #dataloader=SST2Loader() #dataloader=IMDBLoader() -dataloader=yelpLoader(fine_grained=True) +dataloader=YelpLoader(fine_grained=True) datainfo=dataloader.process(ops.datapath,char_level_op=True,split_dev_op=False) char_vocab=ops.char_cnn_config["alphabet"]["en"]["lower"]["alphabet"] ops.number_of_characters=len(char_vocab) diff --git a/reproduction/text_classification/train_dpcnn.py b/reproduction/text_classification/train_dpcnn.py index 70570970..6cce453b 100644 --- a/reproduction/text_classification/train_dpcnn.py +++ b/reproduction/text_classification/train_dpcnn.py @@ -3,15 +3,14 @@ import torch.cuda from fastNLP.core.utils import cache_results from torch.optim import SGD -from torch.optim.lr_scheduler import CosineAnnealingLR, LambdaLR +from torch.optim.lr_scheduler import CosineAnnealingLR from fastNLP.core.trainer import Trainer from fastNLP import CrossEntropyLoss, AccuracyMetric -from fastNLP.modules.encoder.embedding import StaticEmbedding, CNNCharEmbedding, StackEmbedding +from fastNLP.embeddings import StaticEmbedding from reproduction.text_classification.model.dpcnn import DPCNN -from data.yelpLoader import yelpLoader +from fastNLP.io.data_loader import YelpLoader from fastNLP.core.sampler import BucketSampler -import torch.nn as nn -from fastNLP.core import LRScheduler, Callback +from fastNLP.core import LRScheduler from fastNLP.core.const import Const as C from fastNLP.core.vocabulary import VocabularyOption from utils.util_init import set_rng_seeds @@ -59,7 +58,7 @@ print('RNG SEED: {}'.format(ops.seed)) @cache_results(ops.model_dir_or_name+'-data-cache') def load_data(): - datainfo = yelpLoader(fine_grained=True, lower=True).process( + datainfo = YelpLoader(fine_grained=True, lower=True).process( paths=ops.datapath, train_ds=['train'], src_vocab_op=ops.src_vocab_op) for ds in datainfo.datasets.values(): ds.apply_field(len, C.INPUT, C.INPUT_LEN) diff --git a/reproduction/text_classification/train_lstm.py b/reproduction/text_classification/train_lstm.py index 4ecc61a1..40f77061 100644 --- a/reproduction/text_classification/train_lstm.py +++ b/reproduction/text_classification/train_lstm.py @@ -3,20 +3,13 @@ import os os.environ['FASTNLP_BASE_URL'] = 'http://10.141.222.118:8888/file/download/' os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' - -import torch.nn as nn - -from data.IMDBLoader import IMDBLoader -from fastNLP.modules.encoder.embedding import StaticEmbedding +from fastNLP.io.data_loader import IMDBLoader +from fastNLP.embeddings import StaticEmbedding from model.lstm import BiLSTMSentiment -from fastNLP.core.const import Const as C from fastNLP import CrossEntropyLoss, AccuracyMetric -from fastNLP import Trainer, Tester +from fastNLP import Trainer from torch.optim import Adam -from fastNLP.io.model_io import ModelLoader, ModelSaver - -import argparse class Config(): diff --git a/reproduction/text_classification/train_lstm_att.py b/reproduction/text_classification/train_lstm_att.py index a6f0dd03..1052f606 100644 --- a/reproduction/text_classification/train_lstm_att.py +++ b/reproduction/text_classification/train_lstm_att.py @@ -3,20 +3,13 @@ import os os.environ['FASTNLP_BASE_URL'] = 'http://10.141.222.118:8888/file/download/' os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' - -import torch.nn as nn - -from data.IMDBLoader import IMDBLoader -from fastNLP.modules.encoder.embedding import StaticEmbedding +from fastNLP.io.data_loader import IMDBLoader +from fastNLP.embeddings import StaticEmbedding from model.lstm_self_attention import BiLSTM_SELF_ATTENTION -from fastNLP.core.const import Const as C from fastNLP import CrossEntropyLoss, AccuracyMetric -from fastNLP import Trainer, Tester +from fastNLP import Trainer from torch.optim import Adam -from fastNLP.io.model_io import ModelLoader, ModelSaver - -import argparse class Config(): diff --git a/test/core/test_dataset.py b/test/core/test_dataset.py index 0228f207..9c05c334 100644 --- a/test/core/test_dataset.py +++ b/test/core/test_dataset.py @@ -1,4 +1,5 @@ import os +import sys import unittest from fastNLP import DataSet @@ -79,6 +80,16 @@ class TestDataSetMethods(unittest.TestCase): self.assertFalse("x" in dd.field_arrays) self.assertTrue("y" in dd.field_arrays) + def test_delete_instance(self): + dd = DataSet() + old_length = 2 + dd.add_field("x", [[1, 2, 3]] * old_length) + dd.add_field("y", [[1, 2, 3, 4]] * old_length) + dd.delete_instance(0) + self.assertEqual(len(dd), old_length-1) + dd.delete_instance(0) + self.assertEqual(len(dd), old_length-2) + def test_getitem(self): ds = DataSet({"x": [[1, 2, 3, 4]] * 40, "y": [[5, 6]] * 40}) ins_1, ins_0 = ds[0], ds[1] diff --git a/test/core/test_dist_trainer.py b/test/core/test_dist_trainer.py new file mode 100644 index 00000000..c6879634 --- /dev/null +++ b/test/core/test_dist_trainer.py @@ -0,0 +1,167 @@ +import unittest + +import numpy as np +import torch.cuda +from fastNLP import DataSet +from fastNLP import Instance +from fastNLP import CrossEntropyLoss, BCELoss +from fastNLP import SGD +from fastNLP.core.dist_trainer import DistTrainer, get_local_rank +from fastNLP.models.base_model import NaiveClassifier +import shutil +import os +import subprocess +from argparse import ArgumentParser +from fastNLP.core.callback import EchoCallback +from fastNLP import AccuracyMetric + +def prepare_fake_dataset(): + mean = np.array([-3, -3]) + cov = np.array([[1, 0], [0, 1]]) + class_A = np.random.multivariate_normal(mean, cov, size=(1000,)) + + mean = np.array([3, 3]) + cov = np.array([[1, 0], [0, 1]]) + class_B = np.random.multivariate_normal(mean, cov, size=(1000,)) + + data_set = DataSet([Instance(x=[float(item[0]), float(item[1])], y=0) for item in class_A] + + [Instance(x=[float(item[0]), float(item[1])], y=1) for item in class_B]) + return data_set + +def prepare_fake_dataset2(*args, size=100): + ys = np.random.randint(4, size=100, dtype=np.int64) + data = {'y': ys} + for arg in args: + data[arg] = np.random.randn(size, 5) + return DataSet(data=data) + +def set_rng_seed(seed): + np.random.seed(seed) + +def prepare_env(): + def prepare_fake_dataset(): + mean = np.array([-3, -3]) + cov = np.array([[1, 0], [0, 1]]) + class_A = np.random.multivariate_normal(mean, cov, size=(1000,)) + + mean = np.array([3, 3]) + cov = np.array([[1, 0], [0, 1]]) + class_B = np.random.multivariate_normal(mean, cov, size=(1000,)) + + data_set = DataSet([Instance(x=[float(item[0]), float(item[1])], y=[0.0]) for item in class_A] + + [Instance(x=[float(item[0]), float(item[1])], y=[1.0]) for item in class_B]) + return data_set + + data_set = prepare_fake_dataset() + data_set.set_input("x") + data_set.set_target("y") + model = NaiveClassifier(2, 1) + return data_set, model + +class TestDistTrainer(unittest.TestCase): + save_path = './save_cp' + + def run1(self): + # test distributed training + print('local rank', get_local_rank()) + set_rng_seed(100) + data_set = prepare_fake_dataset() + data_set.set_input("x", flag=True) + data_set.set_target("y", flag=True) + + model = NaiveClassifier(2, 2) + + trainer = DistTrainer( + model=model, train_data=data_set, optimizer=SGD(lr=0.1), + loss=CrossEntropyLoss(pred="predict", target="y"), + batch_size_per_gpu=8, n_epochs=3, print_every=50, save_path=self.save_path, + ) + trainer.train() + """ + # 应该正确运行 + """ + if trainer.is_master and os.path.exists(self.save_path): + shutil.rmtree(self.save_path) + + def run2(self): + # test fp16 with distributed training + print('local rank', get_local_rank()) + set_rng_seed(100) + data_set = prepare_fake_dataset() + data_set.set_input("x", flag=True) + data_set.set_target("y", flag=True) + + model = NaiveClassifier(2, 2) + + trainer = DistTrainer( + model=model, train_data=data_set, optimizer=SGD(lr=0.1), + loss=CrossEntropyLoss(pred="predict", target="y"), + batch_size_per_gpu=8, n_epochs=3, print_every=50, save_path=self.save_path, + fp16='O1' + ) + trainer.train() + """ + # 应该正确运行 + """ + if trainer.is_master and os.path.exists(self.save_path): + shutil.rmtree(self.save_path) + + def run3(self): + set_rng_seed(100) + data_set, model = prepare_env() + trainer = DistTrainer( + data_set, model, optimizer=None, + loss=BCELoss(pred="predict", target="y"), + n_epochs=3, print_every=50, + callbacks_all=[EchoCallback('callbacks_all')], + callbacks_master=[EchoCallback('callbacks_master')] + ) + trainer.train() + + def run4(self): + set_rng_seed(100) + data_set, model = prepare_env() + + train_set, dev_set = data_set.split(0.3) + + model = NaiveClassifier(2, 1) + + trainer = DistTrainer( + train_set, model, optimizer=SGD(lr=0.1), + loss=BCELoss(pred="predict", target="y"), + batch_size_per_gpu=32, n_epochs=3, print_every=50, dev_data=dev_set, + metrics=AccuracyMetric(pred="predict", target="y"), validate_every=-1, save_path=None, + ) + trainer.train() + """ + # 应该正确运行 + """ + + def run_dist(self, run_id): + if torch.cuda.is_available(): + ngpu = min(2, torch.cuda.device_count()) + path = __file__ + cmd = ['python', '-m', 'torch.distributed.launch', + '--nproc_per_node', str(ngpu), path, '--test', str(run_id)] + print(' '.join(cmd)) + subprocess.check_call(cmd) + + def test_normal_run(self): + self.run_dist(1) + + def no_test_fp16(self): + self.run_dist(2) + + def test_callback(self): + self.run_dist(3) + + def test_dev_data(self): + self.run_dist(4) + +if __name__ == '__main__': + runner = TestDistTrainer() + parser = ArgumentParser() + parser.add_argument('--test', type=int) + args, _ = parser.parse_known_args() + if args.test and hasattr(runner, 'run%s'%args.test): + getattr(runner, 'run%s'%args.test)() diff --git a/test/core/test_field.py b/test/core/test_field.py index e9053f37..c46e2de2 100644 --- a/test/core/test_field.py +++ b/test/core/test_field.py @@ -170,22 +170,22 @@ class TestFieldArray(unittest.TestCase): def test_append(self): with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append(0) with self.assertRaises(Exception): - fa = FieldArray("y", [1.1, 2.2, 3.3, 4.4, 5.5], is_input=True) + fa = FieldArray("y", [1.1, 2.2, 3.3, 4.4, 5.5], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([1, 2, 3, 4, 5]) with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([]) with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append(["str", 0, 0, 0, 1.89]) - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1.0, 2.0, 3.0, 4.0, 5.0]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1.0, 2.0, 3.0, 4.0, 5.0]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([1.2, 2.3, 3.4, 4.5, 5.6]) self.assertEqual(len(fa), 3) self.assertEqual(fa[2], [1.2, 2.3, 3.4, 4.5, 5.6]) diff --git a/test/core/test_metrics.py b/test/core/test_metrics.py index 9c8a586c..236066d6 100644 --- a/test/core/test_metrics.py +++ b/test/core/test_metrics.py @@ -7,7 +7,7 @@ from fastNLP import AccuracyMetric from fastNLP.core.metrics import _pred_topk, _accuracy_topk from fastNLP.core.vocabulary import Vocabulary from collections import Counter -from fastNLP.core.metrics import SpanFPreRecMetric +from fastNLP.core.metrics import SpanFPreRecMetric, ExtractiveQAMetric def _generate_tags(encoding_type, number_labels=4): @@ -347,3 +347,46 @@ class TestUsefulFunctions(unittest.TestCase): _ = _pred_topk(np.random.randint(0, 3, size=(10, 1))) # 跑通即可 + + +class TestExtractiveQAMetric(unittest.TestCase): + + def test_cast_1(self): + qa_prediction = torch.FloatTensor([[[-0.4424, -0.4579, -0.7376, 1.8129, 0.1316, 1.6566, -1.2169, + -0.3782, 0.8240], + [-1.2348, -0.1876, -0.1462, -0.4834, -0.6692, -0.9735, -1.1563, + -0.3562, -1.4116], + [-1.6550, -0.9555, 0.3782, -1.3160, -1.5835, -0.3443, -1.7858, + -2.0023, 0.0075], + [-0.3772, -0.5447, -1.5631, 1.1614, 1.4598, -1.2764, 0.5186, + 0.3832, -0.1540], + [-0.1011, 0.0600, 1.1090, -0.3545, 0.1284, 1.1484, -1.0120, + -1.3508, -0.9513], + [1.8948, 0.8627, -2.1359, 1.3740, -0.7499, 1.5019, 0.6919, + -0.0842, -0.4294]], + + [[-0.2802, 0.6941, -0.4788, -0.3845, 1.7752, 1.2950, -1.9490, + -1.4138, -0.8853], + [-1.3752, -0.5457, -0.5305, 0.4018, 0.2934, 0.7931, 2.3845, + -1.0726, 0.0364], + [0.3621, 0.2609, 0.1269, -0.5950, 0.7212, 0.5959, 1.6264, + -0.8836, -0.9320], + [0.2003, -1.0758, -1.1560, -0.6472, -1.7549, 0.1264, 0.6044, + -1.6857, 1.1571], + [1.4277, -0.4915, 0.4496, 2.2027, 0.0730, -3.1792, -0.5125, + 3.5837, 1.0184], + [1.6495, 1.7145, -0.2143, -0.1230, -0.2205, 0.8250, 0.4943, + -0.9025, 0.0864]]]) + qa_prediction = qa_prediction.permute(1, 2, 0) + pred1, pred2 = qa_prediction.split(1, dim=-1) + pred1 = pred1.squeeze(-1) + pred2 = pred2.squeeze(-1) + target1 = torch.LongTensor([3, 0, 2, 4, 4, 0]) + target2 = torch.LongTensor([4, 1, 6, 8, 7, 1]) + metric = ExtractiveQAMetric() + metric.evaluate(pred1, pred2, target1, target2) + result = metric.get_metric() + truth = {'EM': 62.5, 'f_1': 72.5, 'noAns-f_1': 50.0, 'noAns-EM': 50.0, 'hasAns-f_1': 95.0, 'hasAns-EM': 75.0} + for k, v in truth.items(): + self.assertTrue(k in result) + self.assertEqual(v, result[k]) diff --git a/test/data_for_tests/sample_mnli.tsv b/test/data_for_tests/sample_mnli.tsv new file mode 100644 index 00000000..9a30b95b --- /dev/null +++ b/test/data_for_tests/sample_mnli.tsv @@ -0,0 +1,12 @@ +index promptID pairID genre sentence1_binary_parse sentence2_binary_parse sentence1_parse sentence2_parse sentence1 sentence2 label1 label2 label3 label4 label5 gold_label +0 63735 63735n slate ( ( The ( new rights ) ) ( are ( nice enough ) ) ) ( Everyone ( really ( likes ( the ( newest benefits ) ) ) ) ) (ROOT (S (NP (DT The) (JJ new) (NNS rights)) (VP (VBP are) (ADJP (JJ nice) (RB enough))))) (ROOT (S (NP (NN Everyone)) (VP (ADVP (RB really)) (VBZ likes) (NP (DT the) (JJS newest) (NNS benefits))))) The new rights are nice enough Everyone really likes the newest benefits neutral entailment neutral neutral neutral neutral +1 91383 91383c government ( ( This site ) ( ( includes ( ( ( ( a list ) ( of ( all ( award winners ) ) ) ) and ) ( ( a ( searchable database ) ) ( of ( Government ( Executive articles ) ) ) ) ) ) . ) ) ( ( ( The ( Government ( Executive articles ) ) ) ( housed ( on ( the website ) ) ) ) ( ( ( are not ) ( able ( to ( be searched ) ) ) ) . ) ) (ROOT (S (NP (DT This) (NN site)) (VP (VBZ includes) (NP (NP (NP (DT a) (NN list)) (PP (IN of) (NP (DT all) (NN award) (NNS winners)))) (CC and) (NP (NP (DT a) (JJ searchable) (NN database)) (PP (IN of) (NP (NNP Government) (NNP Executive) (NNS articles)))))) (. .))) (ROOT (S (NP (NP (DT The) (NNP Government) (NNP Executive) (NNS articles)) (VP (VBN housed) (PP (IN on) (NP (DT the) (NN website))))) (VP (VBP are) (RB not) (ADJP (JJ able) (S (VP (TO to) (VP (VB be) (ADJP (JJ searched))))))) (. .))) This site includes a list of all award winners and a searchable database of Government Executive articles. The Government Executive articles housed on the website are not able to be searched. contradiction contradiction contradiction contradiction contradiction contradiction +2 755 755e telephone ( ( ( ( uh ( i ( ( do n't ) ( know ( ( i i ) ( have ( ( mixed emotions ) ( about ( him ( ( uh sometimes ) ( i ( like him ) ) ) ) ) ) ) ) ) ) ) ) but ) ( ( at ( the ( same times ) ) ) ( i ( love ( to ( see somebody ) ) ) ) ) ) ( beat him ) ) ( I ( ( ( ( ( ( like him ) ( for ( the ( most part ) ) ) ) , ) but ) ( ( would still ) ( enjoy ( seeing ( someone ( beat him ) ) ) ) ) ) . ) ) (ROOT (SINV (S (S (INTJ (UH uh)) (NP (FW i)) (VP (VBP do) (RB n't) (VP (VB know) (NP (NP (FW i) (FW i)) (SBAR (S (VP (VBP have) (VP (VBN mixed) (NP (NNS emotions)) (PP (IN about) (S (NP (PRP him)) (VP (VBG uh) (ADVP (RB sometimes)) (NP (NP (FW i)) (PP (IN like) (NP (PRP him))))))))))))))) (CC but) (S (PP (IN at) (NP (DT the) (JJ same) (NNS times))) (NP (FW i)) (VP (VBP love) (S (VP (TO to) (VP (VB see) (NP (NN somebody)))))))) (VP (VBD beat)) (NP (PRP him)))) (ROOT (S (NP (PRP I)) (VP (VP (VBP like) (NP (PRP him)) (PP (IN for) (NP (DT the) (JJS most) (NN part)))) (, ,) (CC but) (VP (MD would) (ADVP (RB still)) (VP (VB enjoy) (S (VP (VBG seeing) (S (NP (NN someone)) (VP (VB beat) (NP (PRP him))))))))) (. .))) uh i don't know i i have mixed emotions about him uh sometimes i like him but at the same times i love to see somebody beat him I like him for the most part, but would still enjoy seeing someone beat him. entailment entailment entailment entailment entailment entailment +3 78013 78013c telephone ( yeah ( ( i i ) ( think ( ( my ( favorite restaurant ) ) ( ( is always ) ( been ( ( the ( one closest ) ) ( you ( ( know ( the closest ) ) ( ( as long ) ( as ( it ( 's ( it ( meets ( ( the ( minimum criteria ) ) ( you ( know ( of ( good food ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ( ( My ( favorite restaurants ) ) ( ( ( ( are always ) ( ( ( ( ( at least ) a ) hundred ) miles ) away ) ) ( from ( my house ) ) ) . ) ) (ROOT (S (VP (VB yeah) (NP (NP (FW i) (FW i)) (SBAR (S (VP (VBP think) (SBAR (S (NP (PRP$ my) (JJ favorite) (NN restaurant)) (VP (VBZ is) (ADVP (RB always)) (VP (VBN been) (NP (NP (DT the) (CD one) (JJS closest)) (SBAR (S (NP (PRP you)) (VP (VBP know) (NP (DT the) (JJS closest)) (ADVP (ADVP (RB as) (RB long)) (SBAR (IN as) (S (NP (PRP it)) (VP (VBZ 's) (SBAR (S (NP (PRP it)) (VP (VBZ meets) (NP (NP (DT the) (JJ minimum) (NNS criteria)) (SBAR (S (NP (PRP you)) (VP (VBP know) (PP (IN of) (NP (JJ good) (NN food))))))))))))))))))))))))))))) (ROOT (S (NP (PRP$ My) (JJ favorite) (NNS restaurants)) (VP (VBP are) (ADVP (RB always)) (ADVP (NP (QP (IN at) (JJS least) (DT a) (CD hundred)) (NNS miles)) (RB away)) (PP (IN from) (NP (PRP$ my) (NN house)))) (. .))) yeah i i think my favorite restaurant is always been the one closest you know the closest as long as it's it meets the minimum criteria you know of good food My favorite restaurants are always at least a hundred miles away from my house. contradiction contradiction contradiction contradiction contradiction contradiction +4 96377 96377c telephone ( i ( ( do n't ) ( know ( um ( do ( you ( do ( ( a lot ) ( of camping ) ) ) ) ) ) ) ) ) ( I ( ( know exactly ) . ) ) (ROOT (S (NP (FW i)) (VP (VBP do) (RB n't) (VP (VB know) (SBAR (S (NP (NN um)) (VP (VBP do) (SBAR (S (NP (PRP you)) (VP (VBP do) (NP (NP (DT a) (NN lot)) (PP (IN of) (NP (NN camping)))))))))))))) (ROOT (S (NP (PRP I)) (VP (VBP know) (ADVP (RB exactly))) (. .))) i don't know um do you do a lot of camping I know exactly. contradiction contradiction contradiction contradiction contradiction contradiction +5 139749 139749c telephone ( well ( that ( would ( be ( ( a help ) ( i ( wish ( they ( would ( do ( that ( ( ( here ( we ( have ( got ( so ( ( little ( landfill space ) ) ( left ( that ( we ( 're ( going ( to ( ( run out ) ( before ( ( the end ) ( of ( this decade ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) and ) ( it ( ( 's really ) ( going ( to be ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ( We ( ( have ( plenty ( of ( space ( in ( the landfill ) ) ) ) ) ) . ) ) (ROOT (FRAG (ADVP (RB well)) (SBAR (WHNP (WDT that)) (S (VP (MD would) (VP (VB be) (NP (NP (DT a) (NN help)) (SBAR (S (NP (FW i)) (VP (VBP wish) (SBAR (S (NP (PRP they)) (VP (MD would) (VP (VB do) (SBAR (IN that) (S (S (ADVP (RB here)) (NP (PRP we)) (VP (VBP have) (VP (VBN got) (SBAR (IN so) (S (NP (JJ little) (NN landfill) (NN space)) (VP (VBD left) (SBAR (IN that) (S (NP (PRP we)) (VP (VBP 're) (VP (VBG going) (S (VP (TO to) (VP (VB run) (PRT (RP out)) (PP (IN before) (NP (NP (DT the) (NN end)) (PP (IN of) (NP (DT this) (NN decade)))))))))))))))))) (CC and) (S (NP (PRP it)) (VP (VBZ 's) (ADVP (RB really)) (VP (VBG going) (S (VP (TO to) (VP (VB be))))))))))))))))))))))) (ROOT (S (NP (PRP We)) (VP (VBP have) (NP (NP (RB plenty)) (PP (IN of) (NP (NP (NN space)) (PP (IN in) (NP (DT the) (NN landfill))))))) (. .))) well that would be a help i wish they would do that here we have got so little landfill space left that we're going to run out before the end of this decade and it's really going to be We have plenty of space in the landfill. contradiction contradiction contradiction contradiction contradiction contradiction +6 101415 101415c telephone ( yeah ( ( ( i know ) and ) ( i ( did ( that ( ( ( all ( through college ) ) and ) ( it ( worked too ) ) ) ) ) ) ) ) ( I ( ( ( did ( that all ) ) ( through college ) ) ( but ( it ( never worked ) ) ) ) ) (ROOT (S (VP (VB yeah) (S (S (NP (FW i)) (VP (VBP know))) (CC and) (S (NP (FW i)) (VP (VBD did) (SBAR (IN that) (S (S (NP (DT all)) (PP (IN through) (NP (NN college)))) (CC and) (S (NP (PRP it)) (VP (VBD worked) (ADVP (RB too)))))))))))) (ROOT (S (NP (PRP I)) (VP (VBD did) (ADVP (IN that) (DT all)) (PP (IN through) (NP (NN college))) (SBAR (CC but) (S (NP (PRP it)) (ADVP (RB never)) (VP (VBD worked))))))) yeah i know and i did that all through college and it worked too I did that all through college but it never worked contradiction contradiction contradiction contradiction contradiction contradiction +7 93958 93958n travel ( ( ( ( ( Calcutta ( seems ( to ( be ( ( the ( only ( other ( production center ) ) ) ) ( ( having ( any pretensions ) ) ( to ( ( artistic creativity ) ( at all ) ) ) ) ) ) ) ) ) , ) but ) ( ironically ( you ( ( 're actually ) ( ( more ( likely ( to ( see ( ( the works ) ( of ( ( ( Satyajit Ray ) or ) ( ( Mrinal Sen ) ( shown ( in ( Europe ( or ( North America ) ) ) ) ) ) ) ) ) ) ) ) ) ( than ( in ( India itself ) ) ) ) ) ) ) ) . ) ( ( Most ( of ( ( Mrinal ( Sen 's ) ) work ) ) ) ( ( can ( be ( found ( in ( European collections ) ) ) ) ) . ) ) (ROOT (S (S (NP (NNP Calcutta)) (VP (VBZ seems) (S (VP (TO to) (VP (VB be) (NP (NP (DT the) (JJ only) (JJ other) (NN production) (NN center)) (VP (VBG having) (NP (DT any) (NNS pretensions)) (PP (TO to) (NP (NP (JJ artistic) (NN creativity)) (ADVP (IN at) (DT all))))))))))) (, ,) (CC but) (S (ADVP (RB ironically)) (NP (PRP you)) (VP (VBP 're) (ADVP (RB actually)) (ADJP (ADJP (RBR more) (JJ likely) (S (VP (TO to) (VP (VB see) (NP (NP (DT the) (NNS works)) (PP (IN of) (NP (NP (NNP Satyajit) (NNP Ray)) (CC or) (NP (NP (NNP Mrinal) (NNP Sen)) (VP (VBN shown) (PP (IN in) (NP (NNP Europe) (CC or) (NNP North) (NNP America)))))))))))) (ADVP (IN than) (PP (IN in) (S (VP (VBG India) (NP (PRP itself))))))))) (. .))) (ROOT (S (NP (NP (JJS Most)) (PP (IN of) (NP (NP (NNP Mrinal) (NNP Sen) (POS 's)) (NN work)))) (VP (MD can) (VP (VB be) (VP (VBN found) (PP (IN in) (NP (JJ European) (NNS collections)))))) (. .))) Calcutta seems to be the only other production center having any pretensions to artistic creativity at all, but ironically you're actually more likely to see the works of Satyajit Ray or Mrinal Sen shown in Europe or North America than in India itself. Most of Mrinal Sen's work can be found in European collections. neutral neutral entailment neutral neutral neutral +8 12567 12567c slate ( ( If ( ( that investor ) ( were ( willing ( to ( pay ( extra ( for ( ( the security ) ( of ( limited downside ) ) ) ) ) ) ) ) ) ) ) ( , ( she ( ( could ( ( buy ( put options ) ) ( with ( ( a ( strike price ) ) ( of ( ( ( $ 98 ) , ) ( which ( would ( ( ( lock ( in ( ( her profit ) ( on ( ( the shares ) ( at ( $ 18 ) ) ) ) ) ) ) , ) ( less ( whatever ( ( the options ) cost ) ) ) ) ) ) ) ) ) ) ) ) . ) ) ) ) ( ( THe ( strike price ) ) ( ( could ( be ( $ 8 ) ) ) . ) ) (ROOT (S (SBAR (IN If) (S (NP (DT that) (NN investor)) (VP (VBD were) (ADJP (JJ willing) (S (VP (TO to) (VP (VB pay) (NP (NP (JJ extra)) (PP (IN for) (NP (NP (DT the) (NN security)) (PP (IN of) (NP (JJ limited) (NN downside))))))))))))) (, ,) (NP (PRP she)) (VP (MD could) (VP (VB buy) (NP (NN put) (NNS options)) (PP (IN with) (NP (NP (DT a) (NN strike) (NN price)) (PP (IN of) (NP (NP ($ $) (CD 98)) (, ,) (SBAR (WHNP (WDT which)) (S (VP (MD would) (VP (VB lock) (PP (IN in) (NP (NP (PRP$ her) (NN profit)) (PP (IN on) (NP (NP (DT the) (NNS shares)) (PP (IN at) (NP ($ $) (CD 18))))))) (, ,) (ADVP (ADVP (RBR less)) (SBAR (WHNP (WDT whatever)) (S (NP (DT the) (NNS options)) (VP (VBD cost))))))))))))))) (. .))) (ROOT (S (NP (NNP THe) (NN strike) (NN price)) (VP (MD could) (VP (VB be) (NP ($ $) (CD 8)))) (. .))) If that investor were willing to pay extra for the security of limited downside, she could buy put options with a strike price of $98, which would lock in her profit on the shares at $18, less whatever the options cost. THe strike price could be $8. contradiction contradiction contradiction contradiction contradiction contradiction +9 117487 117487n slate ( ( 3 -RRB- ) ( ( Dare ( you ( ( ( rise ( to ( ( ( ( the occasion ) , ) ( like Raskolnikov ) ) , ) ) ) and ) ( reject ( ( the ( petty rules ) ) ( that ( govern ( lesser men ) ) ) ) ) ) ) ) ? ) ) ( ( ( Would you ) ( ( ( rise up ) and ) ( defeaat ( ( all ( evil lords ) ) ( in ( the town ) ) ) ) ) ) ? ) (ROOT (S (LST (LS 3) (-RRB- -RRB-)) (VP (VB Dare) (S (NP (PRP you)) (VP (VP (VB rise) (PP (TO to) (NP (NP (DT the) (NN occasion)) (, ,) (PP (IN like) (NP (NNP Raskolnikov))) (, ,)))) (CC and) (VP (VB reject) (NP (NP (DT the) (JJ petty) (NNS rules)) (SBAR (WHNP (WDT that)) (S (VP (VBP govern) (NP (JJR lesser) (NNS men)))))))))) (. ?))) (ROOT (SQ (MD Would) (NP (PRP you)) (VP (VP (VB rise) (PRT (RP up))) (CC and) (VP (VB defeaat) (NP (NP (DT all) (JJ evil) (NNS lords)) (PP (IN in) (NP (DT the) (NN town)))))) (. ?))) 3) Dare you rise to the occasion, like Raskolnikov, and reject the petty rules that govern lesser men? Would you rise up and defeaat all evil lords in the town? neutral neutral neutral neutral neutral neutral +10 9616 9616c travel ( ( The ( ( most important ) directions ) ) ( ( ( are ( simply ( ( up and ) up ) ) ) ( ( ( ( ( ( ( ( leads eventually ) ( to ( the cathedral ) ) ) and ) ( fortress ( commanding ( the hilltop ) ) ) ) , ) and ) down ) ( inevitably ( ( leads ( to ( one ( of ( three gates ) ) ) ) ) ( through ( ( the wall ) ( to ( the ( new town ) ) ) ) ) ) ) ) ) . ) ) ( Go ( ( downwards ( to ( one ( of ( ( ( the gates ) , ) ( ( all ( of which ) ) ( will ( ( lead you ) ( into ( the cathedral ) ) ) ) ) ) ) ) ) ) . ) ) (ROOT (S (NP (DT The) (ADJP (RBS most) (JJ important)) (NNS directions)) (VP (VBP are) (PRN (ADVP (RB simply)) (ADVP (RB up) (CC and) (RB up))) (VP (VP (VBZ leads) (ADVP (RB eventually)) (PP (TO to) (NP (DT the) (NN cathedral)))) (CC and) (VP (VBZ fortress) (NP (JJ commanding) (DT the) (NN hilltop))) (, ,) (CC and) (ADVP (RB down)) (VP (ADVP (RB inevitably)) (VBZ leads) (PP (TO to) (NP (NP (CD one)) (PP (IN of) (NP (CD three) (NNS gates))))) (PP (IN through) (NP (NP (DT the) (NN wall)) (PP (TO to) (NP (DT the) (JJ new) (NN town)))))))) (. .))) (ROOT (S (NP (NNP Go)) (VP (VBZ downwards) (PP (TO to) (NP (NP (CD one)) (PP (IN of) (NP (NP (DT the) (NNS gates)) (, ,) (SBAR (WHNP (DT all) (WHPP (IN of) (WHNP (WDT which)))) (S (VP (MD will) (VP (VB lead) (NP (PRP you)) (PP (IN into) (NP (DT the) (NN cathedral)))))))))))) (. .))) The most important directions are simply up and up leads eventually to the cathedral and fortress commanding the hilltop, and down inevitably leads to one of three gates through the wall to the new town. Go downwards to one of the gates, all of which will lead you into the cathedral. contradiction contradiction entailment contradiction contradiction contradiction diff --git a/test/embeddings/test_char_embedding.py b/test/embeddings/test_char_embedding.py new file mode 100644 index 00000000..ceafe4f5 --- /dev/null +++ b/test/embeddings/test_char_embedding.py @@ -0,0 +1,26 @@ +import unittest + +import torch + +from fastNLP import Vocabulary, DataSet, Instance +from fastNLP.embeddings.char_embedding import LSTMCharEmbedding, CNNCharEmbedding + + +class TestCharEmbed(unittest.TestCase): + def test_case_1(self): + ds = DataSet([Instance(words=['hello', 'world']), Instance(words=['Jack'])]) + vocab = Vocabulary().from_dataset(ds, field_name='words') + self.assertEqual(len(vocab), 5) + embed = LSTMCharEmbedding(vocab, embed_size=60) + x = torch.LongTensor([[2, 1, 0], [4, 3, 4]]) + y = embed(x) + self.assertEqual(tuple(y.size()), (2, 3, 60)) + + def test_case_2(self): + ds = DataSet([Instance(words=['hello', 'world']), Instance(words=['Jack'])]) + vocab = Vocabulary().from_dataset(ds, field_name='words') + self.assertEqual(len(vocab), 5) + embed = CNNCharEmbedding(vocab, embed_size=60) + x = torch.LongTensor([[2, 1, 0], [4, 3, 4]]) + y = embed(x) + self.assertEqual(tuple(y.size()), (2, 3, 60)) diff --git a/test/embeddings/test_stack_embeddings.py b/test/embeddings/test_stack_embeddings.py new file mode 100644 index 00000000..2eb0b414 --- /dev/null +++ b/test/embeddings/test_stack_embeddings.py @@ -0,0 +1,20 @@ +import unittest + +import torch + +from fastNLP import Vocabulary, DataSet, Instance +from fastNLP.embeddings import LSTMCharEmbedding, CNNCharEmbedding, StackEmbedding + + +class TestCharEmbed(unittest.TestCase): + def test_case_1(self): + ds = DataSet([Instance(words=['hello', 'world']), Instance(words=['hello', 'Jack'])]) + vocab = Vocabulary().from_dataset(ds, field_name='words') + self.assertEqual(len(vocab), 5) + cnn_embed = CNNCharEmbedding(vocab, embed_size=60) + lstm_embed = LSTMCharEmbedding(vocab, embed_size=70) + embed = StackEmbedding([cnn_embed, lstm_embed]) + x = torch.LongTensor([[2, 1, 0], [4, 3, 4]]) + y = embed(x) + self.assertEqual(tuple(y.size()), (2, 3, 130)) + diff --git a/test/embeddings/test_static_embedding.py b/test/embeddings/test_static_embedding.py new file mode 100644 index 00000000..0c8fc739 --- /dev/null +++ b/test/embeddings/test_static_embedding.py @@ -0,0 +1,15 @@ +import unittest + +from fastNLP.embeddings import StaticEmbedding +from fastNLP import Vocabulary +import torch + +class TestRandomSameEntry(unittest.TestCase): + def test_same_vector(self): + vocab = Vocabulary().add_word_lst(["The", "the", "THE"]) + embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5, lower=True) + words = torch.LongTensor([[vocab.to_index(word) for word in ["The", "the", "THE"]]]) + words = embed(words) + embed_0 = words[0, 0] + for i in range(1, words.size(1)): + assert torch.sum(embed_0==words[0, i]).eq(len(embed_0)) diff --git a/test/io/test_data_loader.py b/test/io/test_data_loader.py new file mode 100644 index 00000000..5b1bb749 --- /dev/null +++ b/test/io/test_data_loader.py @@ -0,0 +1,15 @@ +import unittest + +from fastNLP.core.const import Const +from fastNLP.io.data_loader import MNLILoader + + +class TestDataLoader(unittest.TestCase): + + def test_mnli_loader(self): + ds = MNLILoader().process('test/data_for_tests/sample_mnli.tsv', + to_lower=True, get_index=True, seq_len_type='mask') + self.assertTrue('train' in ds.datasets) + self.assertTrue(len(ds.datasets) == 1) + self.assertTrue(len(ds.datasets['train']) == 11) + self.assertTrue(isinstance(ds.datasets['train'][0][Const.INPUT_LENS(0)], list)) diff --git a/test/io/test_embed_loader.py b/test/io/test_embed_loader.py index ff8ecfcf..bbfe8858 100644 --- a/test/io/test_embed_loader.py +++ b/test/io/test_embed_loader.py @@ -16,7 +16,7 @@ class TestEmbedLoader(unittest.TestCase): self.assertEqual(g_m.shape, (4, 50)) w_m = EmbedLoader.load_with_vocab(word2vec, vocab, normalize=True) self.assertEqual(w_m.shape, (4, 50)) - self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 4) + self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 4, delta=1e-4) def test_load_without_vocab(self): words = ['the', 'of', 'in', 'a', 'to', 'and'] @@ -28,13 +28,13 @@ class TestEmbedLoader(unittest.TestCase): self.assertIn(word, vocab) w_m, vocab = EmbedLoader.load_without_vocab(word2vec, normalize=True) self.assertEqual(w_m.shape, (8, 50)) - self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 8) + self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 8, delta=1e-4) for word in words: self.assertIn(word, vocab) # no unk w_m, vocab = EmbedLoader.load_without_vocab(word2vec, normalize=True, unknown=None) self.assertEqual(w_m.shape, (7, 50)) - self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 7) + self.assertAlmostEqual(np.linalg.norm(w_m, axis=1).sum(), 7, delta=1e-4) for word in words: self.assertIn(word, vocab) diff --git a/test/models/test_bert.py b/test/models/test_bert.py index 38a16f9b..05ee6d5a 100644 --- a/test/models/test_bert.py +++ b/test/models/test_bert.py @@ -8,7 +8,7 @@ from fastNLP.models.bert import * class TestBert(unittest.TestCase): def test_bert_1(self): from fastNLP.core.const import Const - from fastNLP.modules.encoder._bert import BertConfig + from fastNLP.modules.encoder.bert import BertConfig model = BertForSequenceClassification(2, BertConfig(32000)) @@ -23,7 +23,7 @@ class TestBert(unittest.TestCase): def test_bert_2(self): from fastNLP.core.const import Const - from fastNLP.modules.encoder._bert import BertConfig + from fastNLP.modules.encoder.bert import BertConfig model = BertForMultipleChoice(2, BertConfig(32000)) @@ -38,7 +38,7 @@ class TestBert(unittest.TestCase): def test_bert_3(self): from fastNLP.core.const import Const - from fastNLP.modules.encoder._bert import BertConfig + from fastNLP.modules.encoder.bert import BertConfig model = BertForTokenClassification(7, BertConfig(32000)) @@ -53,7 +53,7 @@ class TestBert(unittest.TestCase): def test_bert_4(self): from fastNLP.core.const import Const - from fastNLP.modules.encoder._bert import BertConfig + from fastNLP.modules.encoder.bert import BertConfig model = BertForQuestionAnswering(BertConfig(32000)) diff --git a/test/models/test_snli.py b/test/models/test_snli.py new file mode 100644 index 00000000..7a588a4c --- /dev/null +++ b/test/models/test_snli.py @@ -0,0 +1,9 @@ +import unittest +from .model_runner import * +from fastNLP.models.snli import ESIM + + +class TestSNLIModel(unittest.TestCase): + def test_snli(self): + model = ESIM((VOCAB_SIZE, 10), num_labels=NUM_CLS, dropout_rate=0) + RUNNER.run_model_with_task(NLI, model) diff --git a/test/modules/encoder/test_bert.py b/test/modules/encoder/test_bert.py index 2a799478..0fcf01e4 100644 --- a/test/modules/encoder/test_bert.py +++ b/test/modules/encoder/test_bert.py @@ -8,7 +8,7 @@ from fastNLP.models.bert import BertModel class TestBert(unittest.TestCase): def test_bert_1(self): - from fastNLP.modules.encoder._bert import BertConfig + from fastNLP.modules.encoder.bert import BertConfig config = BertConfig(32000) model = BertModel(config)
fastNLP.modules 实现了用于搭建神经网络模型的诸多组件
fastNLP.embeddings 实现了将序列index转为向量序列的功能，包括读取预训练embedding等
fastNLP.io 实现了读写功能，包括数据读入，模型读写等