There are 4 different models in this repository. To train and evaluate them, it is neccessary to assign environment variables DATA_DIR, OUT_DIR and EMBEDDINGS for different file paths and specify config.py files. You then need to execute train.py and evaluate.py files. For all the models, DATA_DIR = data and EMBEDDINGS = embeddings.

For models 1 and 2:

• Training: softmax/scripts/train.py
• Evaluation: softmax/scripts/evaluate.py

• OUT_DIR = softmax/emb_tag_sum/output
• configuration: softmax/emb_tag_sum/config.py

• OUT_DIR = softmax/emb_cat_sum/output
• configuration: softmax/emb_cat_sum/config.py

For models 3 and 4:

• Training: seq2seq/scripts/train.py
• Evaluation: seq2seq/scripts/evaluate.py

• OUT_DIR = seq2seq/emb_tag_sum/output
• configuration: seq2seq/emb_tag_sum/config.py

• OUT_DIR = seq2seq/emb_cat_sum/output
• configuration: seq2seq/emb_cat_sum/config.py

Here's an example of a bash script for training and evaluating model 1:

#!/bin/bash

set -e

export DATA_DIR=data

export OUT_DIR=softmax/emb_tag_sum/output

export EMBEDDINGS_DIR=embeddings

python softmax/scripts/train.py --config softmax/emb_tag_sum/config.py

python softmax/scripts/evaluate.py --test --config softmax/emb_tag_sum/config.py