Sorting Numbers

This is a simple tutorial for sorting numbers with Pointer Networks.

1. Training and validation data should be formatted as below, numbers and their sorted indices separated by \t:

1 2 3 \t 0 1 2
1 3 2 \t 0 2 1
2 1 3 \t 1 0 2
2 3 1 \t 1 2 0
3 1 2 \t 2 0 1
3 2 1 \t 2 1 0
...

2. Modify the type and dimension of the word embedding layer in parameters.py, for exmaple:

UNIT = "word"
EMBED = {"rnn": 100}

4. Run prepare.py to make CSV and index files.

python3 ../prepare.py train

5. Train your model. You can modify the hyperparameters in parameters.py.

python3 ../train.py model train.char_to_idx train.word_to_idx train.csv valid N

6. Predict and evaluate your model.

python3 predict.py model.epochN train.char_to_idx train.word_to_idx test
python3 evaluate.py model.epochN train.char_to_idx train.word_to_idx test