Preparing training data

The training code expects data to be in TFRecord format. data_prep/convert.py can be used to convert text data into this form. An example invocation on the sentiment data from Shen et al. is as below:

python data_prep/convert.py data/sentiment.test.0,data/sentiment.test.1 <vocab file> <output dir>

Training and evaluation

The scripts folder has bash scripts to train and sample from the models.