hummingbird
/
fastNLP
Not watched
2
0
Fork 0
Code
Releases 13 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
266 Commits
3 Branches
25 MB
139 Download
 
 
 
 
 
Tree: bf5d1c347b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'bf5d1c347b'
${ noResults }
fastNLP/reproduction/Char-aware_NLM
FengZiYjun 7514be6f30 - add validation loss into trainer.train 7 years ago
..
LICENSE - add validation loss into trainer.train 7 years ago
README.md - add validation loss into trainer.train 7 years ago
__init__.py - add validation loss into trainer.train 7 years ago
model.py - add validation loss into trainer.train 7 years ago
test.py - add validation loss into trainer.train 7 years ago
test.txt - add validation loss into trainer.train 7 years ago
train.py - add validation loss into trainer.train 7 years ago
train.txt - add validation loss into trainer.train 7 years ago
utilities.py - add validation loss into trainer.train 7 years ago
valid.txt - add validation loss into trainer.train 7 years ago

README.md

PyTorch-Character-Aware-Neural-Language-Model

This is the PyTorch implementation of character-aware neural language model proposed in this paper by Yoon Kim.

Requiredments

The code is run and tested with Python 3.5.2 and PyTorch 0.3.1.

HyperParameters

HyperParam value
LSTM batch size 20
LSTM sequence length 35
LSTM hidden units 300
epochs 35
initial learning rate 1.0
character embedding dimension 15

Demo

Train the model with split train/valid/test data.

python train.py

The trained model will saved in cache/net.pkl.
Test the model.

python test.py

Best result on test set:
PPl=127.2163
cross entropy loss=4.8459

Acknowledgement

This implementation borrowed ideas from

https://github.com/jarfo/kchar

https://github.com/cronos123/Character-Aware-Neural-Language-Models

Delete task

Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.

cancel
confirm

一款轻量级的自然语言处理（NLP）工具包，目标是减少用户项目中的工程型代码，例如数据处理循环、训练循环、多卡运行等

Python Jupyter Notebook Text CSV Markdown