Rational Language Model

This repo contains code for a rational language model

Still needs to be cleaned up a bit. But currently what works is:

• train_pl.py (Used to pretrain the model )
• train_prediction_rational_pl.py (Train the model)

Currently the gumbell softmax is used for training. As this was easiest to implement.

Training

LSTM language model

The training of the Rational Extractor consists of different steps. First we need to train the language model:

1) python train_language_model.py configs/simple_lm_config.yml