Workflow:

1. Manually remove data-bin and checkpoints.

2. Preprocess by running the following:

   $ bash preprocess.sh spa.txt
   

3. Manually assure that all files don't have empty lines.

4. Train by running:

   $ bash train.sh -a <EncoderEmbedDim> -b <DecoderEmbedDim> -c <DecoderOutEmbedDim> -d <EncoderHiddenSize> -e <DecoderHiddenSize>
   

5. Generate predictions by running:

   $ bash generate.sh <predictions_DATE.txt>
   

6. Run an error analysis by running:

   $ python error_analysis.py <predictions_DATE.txt> <errors_DATE.txt>