rajevv / infersent Goto Github PK

The contents and the directory structure to be followed is as follows:

eval_logs/
Models/
    awe/
    default/  -----> stands for LSTM
    bilstm/
    bilstmpooling
models/
runs/
snli_1.0/
data_utils.py
eval.py
glove.840B.300d.txt
glove.840B.300d_aligned.pt
predict.py
rajeev3.yml
senteval.py
train.py
vocab.json

Models: The trained models. 
glove.840B.300d_aligned.pt: the glove vectors aligned with the vocabulary corresponding to the data.

*The following above two have to downloaded from:
https://drive.google.com/drive/folders/1dfbPdJ5WzBkWPMuymy4a7ziFQ8DzZW_0?usp=sharing

Though, no other alternative libraries have not been used. So, the code should run on any regular Pytorch environment. Still, the environment corresponding to the code is provided in rajeev3.yml file.

eval_logs: evaluation logs for all the models
runs: Tensorboard logs. 

The main scripts are:
train.py: To train the models. Please use train.py -h to see the arguments.
eval.py: To evaluate the models on validation and test data. Again, please follow the eval.py -h to see the args.
predict.py: This script takes encoder type, and premise-hypothesis pair to predict the relation between them.