/Sentence_Classifier_CNN

Pytorch Implementation of Sentence Classification Model Using Convolution Neural Networks

Primary LanguagePythonMIT LicenseMIT

Sentence_Classifier_CNN

Shindong Lee

Korea University

Overview

This is Pytorch implementation of the paper "Convolutional Neural Networks for Sentence Classification" by Kim Yoon

Performance

Accuracy on MR dataset (%)

This Implementation Kim's Paper
CNN-rand 79.8 76.1
CNN-static 82.1 81.0
CNN-non-static 81.3 81.5
CNN-multichannel 81.3 81.1

This Implementation is compatible with both CPU and GPU

With GPU RTX 2080ti,

Time for preprocessing < 1 minute
Time for training 1 epoch < 1 second
Time for validation < 1 second

Reference

"Convolutional Neural Networks for Sentence Classification" by Kim Yoon
(paper: https://arxiv.org/pdf/1408.5882.pdf)
(original implementation by Kim Yoon with Theano: https://github.com/yoonkim/CNN_sentence)

Movie Reivew Dataset: (It's already inside the repo)
(http://www.cs.cornell.edu/People/pabo/movie-review-data/rt-polaritydata.tar.gz)

Difference From Paper

Paper used Adadelta, but I used Adam Optimizer

How to Run

  1. download GoogleNews-vectors-negative300.bin from (https://github.com/mmihaltz/word2vec-GoogleNews-vectors)
    or from (https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit)

  2. unzip and save it into "pretrained" directory

  3. run following code

$ python train.py --model [model] --num_epochs [num_epochs] --learning_rate [learning_rate] --mini_batch_size [mini_batch_size]

ex)

$ python train.py --model CNN-rand --num_epochs 50 --learning_rate 0.0001 --mini_batch_size 50

or simply, 

$ python train.py --model CNN-rand

model should be one of [CNN-rand, CNN-static, CNN-non-static, CNN-multichannel]

If you want to resample test set from MR dataset, just remove 4 files [ train_data_pos.txt, train_data_neg.txt, test_data_pos.txt, test_data_neg.txt ] and run the code.