Implementaion of Enhanced LSTM for Natural Language Inference with Tensorflow

"Enhanced LSTM for Natural Language Inference" Qian Chen, Xiaodan Zhu, Zhenhua Ling, Si Wei, Hui Jiang, Diana Inkpen. ACL (2017)

Dependencies

To run it perfectly, you will need:

• Python 2.7
• Tensorflow (preferable 1.8)

Running the Script

1. Download and preprocess the data set

cd data
bash fetch_and_preprocess.sh

2. Train and test model for ESIM

cd scripts/ESIM/
bash train.sh

The results are shown in log.txt and log_result.txt file.

Note

There are three main files in the scripts folder

scripts/main.py

main.py is only compatible when tensorflow version >= 1.8. It uses the latest tf.contrib.cudnn_rnn.CudnnLSTM library that gives you the boosting performance when training the model.

scripts/main1.py

main1.py are implemented the same as main.py but works with lower version of tensorflow.

scripts/main_old.py

main_old.py uses the old libraries, it's slightly slower but gives you stable results compared with the other two mentioned above