cstghitpku/cnn-lstm-bilstm-deepcnn-clstm-in-pytorch

In pytorch Learing Neural Networks likes CNN(Convolutional Neural Networks for Sentence Classification (Y.Kim, EMNLP 2014) 、LSTM、BiLSTM、DeepCNN 、CLSTM、CNN and LSTM

Introduction

• A classification task implement in pytorch, contains some neural networks in models.

• Recenely, I've readjusted the code structure to make it easier to read.
  • The old version in the branch of old-version-17, It still works directly.
  • The newest version in here(master), It also can works directly now.

• This is a version of my own architecture --- pytorch-text-classification .

Requirement

pyorch : 0.3.1
python : 3.6.1
torchtext: 0.2.1
cuda : 8.0 (support cuda speed up, can chose)

Usage

modify the config file, see the Config directory(here) for detail.

1、python main.py
2、python main.py --config_file ./Config/config.cfg 

Model

Contains some neural networks implement in pytorch, see the models for detail.

Data

SST-1 and SST-2.

Result

I haven't adjusted the hyper-parameters seriously, you can also see train log in here.

The following test set accuracy are based on the best dev set accuracy.

Data/Model	% SST-1	% SST-2
CNN-Rand	44.1629	82.2076
CNN	45.4751	84.3453
Bi-LSTM	48.1448	86.8753
Bi-GRU	48.1448	87.3696

Reference

• 基于pytorch的CNN-LSTM神经网络模型调参小结
• Convolutional Neural Networks for Sentence Classification
• Context-Sensitive Lexicon Features for Neural Sentiment Analysis

Question

• if you have any question, you can open a issue or email bamtercelboo@{gmail.com, 163.com}.

• if you have any good suggestions, you can PR or email me.