diff --git a/reproduction/text_classification/README.md b/reproduction/text_classification/README.md index b058fbb2..08c893b7 100644 --- a/reproduction/text_classification/README.md +++ b/reproduction/text_classification/README.md @@ -1,22 +1,28 @@ # text_classification任务模型复现 这里使用fastNLP复现以下模型： + char_cnn :论文链接[Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626v3.pdf) + dpcnn:论文链接[Deep Pyramid Convolutional Neural Networks for TextCategorization](https://ai.tencent.com/ailab/media/publications/ACL3-Brady.pdf) + HAN:论文链接[Hierarchical Attention Networks for Document Classification](https://www.cs.cmu.edu/~diyiy/docs/naacl16.pdf) + +LSTM+self_attention:论文链接[A Structured Self-attentive Sentence Embedding]() + +AWD-LSTM:论文链接[Regularizing and Optimizing LSTM Language Models]() + #待补充 -awd_lstm: -lstm_self_attention(BCN?): -awd-sltm: # 数据集及复现结果汇总 使用fastNLP复现的结果vs论文汇报结果(/前为fastNLP实现，后面为论文报道,-表示论文没有在该数据集上列出结果) -model name | yelp_p | sst-2|IMDB| -:---: | :---: | :---: | :---: -char_cnn | 93.80/95.12 | - |- | -dpcnn | 95.50/97.36 | - |- | -HAN |- | - |-| -BCN| - |- |-| -awd-lstm| - |- |-| +model name | yelp_p | yelp_f | sst-2|IMDB +:---: | :---: | :---: | :---: |----- +char_cnn | 93.80/95.12 | - | - |- +dpcnn | 95.50/97.36 | - | - |- +HAN |- | - | - |- +LSTM| 95.74/- |- |- |88.52/- +AWD-LSTM| 95.96/- |- |- |88.91/- +LSTM+self_attention| 96.34/- | - | - |89.53/- diff --git a/reproduction/text_classification/results_LSTM.xlsx b/reproduction/text_classification/results_LSTM.xlsx deleted file mode 100644 index 0d7b841b..00000000 Binary files a/reproduction/text_classification/results_LSTM.xlsx and /dev/null differ diff --git a/reproduction/text_classification/train_awdlstm.py b/reproduction/text_classification/train_awdlstm.py index ce3e52bc..007b2910 100644 --- a/reproduction/text_classification/train_awdlstm.py +++ b/reproduction/text_classification/train_awdlstm.py @@ -8,9 +8,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' import torch.nn as nn -from data.SSTLoader import SSTLoader from data.IMDBLoader import IMDBLoader -from data.yelpLoader import yelpLoader from fastNLP.modules.encoder.embedding import StaticEmbedding from model.awd_lstm import AWDLSTMSentiment @@ -35,24 +33,15 @@ class Config(): task_name = "IMDB" datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"} - load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51" save_model_path="./result_IMDB_test/" -opt=Config +opt=Config() -# load data -dataloaders = { - "IMDB":IMDBLoader(), - "YELP":yelpLoader(), - "SST-5":SSTLoader(subtree=True,fine_grained=True), - "SST-3":SSTLoader(subtree=True,fine_grained=False) -} - -if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]: - raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']") -dataloader = dataloaders[opt.task_name] +# load data +dataloader=IMDBLoader() datainfo=dataloader.process(opt.datapath) + # print(datainfo.datasets["train"]) # print(datainfo) @@ -71,32 +60,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T def train(datainfo, model, optimizer, loss, metrics, opt): trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss, - metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1, + metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1, n_epochs=opt.train_epoch, save_path=opt.save_model_path) trainer.train() -def test(datainfo, metrics, opt): - # load model - model = ModelLoader.load_pytorch_model(opt.load_model_path) - print("model loaded!") - - # Tester - tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0) - acc = tester.test() - print("acc=",acc) - - - -parser = argparse.ArgumentParser() -parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model') - - -args = parser.parse_args() -if args.mode == 'train': +if __name__ == "__main__": train(datainfo, model, optimizer, loss, metrics, opt) -elif args.mode == 'test': - test(datainfo, metrics, opt) -else: - print('no mode specified for model!') - parser.print_help() diff --git a/reproduction/text_classification/train_lstm.py b/reproduction/text_classification/train_lstm.py index b320e79c..4ecc61a1 100644 --- a/reproduction/text_classification/train_lstm.py +++ b/reproduction/text_classification/train_lstm.py @@ -6,9 +6,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' import torch.nn as nn -from data.SSTLoader import SSTLoader from data.IMDBLoader import IMDBLoader -from data.yelpLoader import yelpLoader from fastNLP.modules.encoder.embedding import StaticEmbedding from model.lstm import BiLSTMSentiment @@ -32,24 +30,15 @@ class Config(): task_name = "IMDB" datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"} - load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51" save_model_path="./result_IMDB_test/" -opt=Config +opt=Config() -# load data -dataloaders = { - "IMDB":IMDBLoader(), - "YELP":yelpLoader(), - "SST-5":SSTLoader(subtree=True,fine_grained=True), - "SST-3":SSTLoader(subtree=True,fine_grained=False) -} - -if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]: - raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']") -dataloader = dataloaders[opt.task_name] +# load data +dataloader=IMDBLoader() datainfo=dataloader.process(opt.datapath) + # print(datainfo.datasets["train"]) # print(datainfo) @@ -68,32 +57,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T def train(datainfo, model, optimizer, loss, metrics, opt): trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss, - metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1, + metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1, n_epochs=opt.train_epoch, save_path=opt.save_model_path) trainer.train() -def test(datainfo, metrics, opt): - # load model - model = ModelLoader.load_pytorch_model(opt.load_model_path) - print("model loaded!") - - # Tester - tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0) - acc = tester.test() - print("acc=",acc) - - - -parser = argparse.ArgumentParser() -parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model') - - -args = parser.parse_args() -if args.mode == 'train': - train(datainfo, model, optimizer, loss, metrics, opt) -elif args.mode == 'test': - test(datainfo, metrics, opt) -else: - print('no mode specified for model!') - parser.print_help() +if __name__ == "__main__": + train(datainfo, model, optimizer, loss, metrics, opt) \ No newline at end of file diff --git a/reproduction/text_classification/train_lstm_att.py b/reproduction/text_classification/train_lstm_att.py index 8db27d09..a6f0dd03 100644 --- a/reproduction/text_classification/train_lstm_att.py +++ b/reproduction/text_classification/train_lstm_att.py @@ -6,9 +6,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches' import torch.nn as nn -from data.SSTLoader import SSTLoader from data.IMDBLoader import IMDBLoader -from data.yelpLoader import yelpLoader from fastNLP.modules.encoder.embedding import StaticEmbedding from model.lstm_self_attention import BiLSTM_SELF_ATTENTION @@ -34,24 +32,15 @@ class Config(): task_name = "IMDB" datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"} - load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51" save_model_path="./result_IMDB_test/" -opt=Config +opt=Config() -# load data -dataloaders = { - "IMDB":IMDBLoader(), - "YELP":yelpLoader(), - "SST-5":SSTLoader(subtree=True,fine_grained=True), - "SST-3":SSTLoader(subtree=True,fine_grained=False) -} - -if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]: - raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']") -dataloader = dataloaders[opt.task_name] +# load data +dataloader=IMDBLoader() datainfo=dataloader.process(opt.datapath) + # print(datainfo.datasets["train"]) # print(datainfo) @@ -70,32 +59,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T def train(datainfo, model, optimizer, loss, metrics, opt): trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss, - metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1, + metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1, n_epochs=opt.train_epoch, save_path=opt.save_model_path) trainer.train() -def test(datainfo, metrics, opt): - # load model - model = ModelLoader.load_pytorch_model(opt.load_model_path) - print("model loaded!") - - # Tester - tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0) - acc = tester.test() - print("acc=",acc) - - - -parser = argparse.ArgumentParser() -parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model') - - -args = parser.parse_args() -if args.mode == 'train': +if __name__ == "__main__": train(datainfo, model, optimizer, loss, metrics, opt) -elif args.mode == 'test': - test(datainfo, metrics, opt) -else: - print('no mode specified for model!') - parser.print_help()