pytorch_seq2seq-rl

A PyTorch template of seq2seq with RL training

This is a toy task of seq2seq which reversing the input list

[5, 6, 4, 6, 7] -> [7, 6, 4, 6, 5]
[1, 3, 7, 8, 8, 9, 9] -> [9, 9, 8, 8, 7, 3, 1]

Usage

• seq2seq

Dataset/generator.ipynb
model_seq2seq.ipynb
model_seq2seq-rl.ipynb

• MNIST (classification)

mnist_ce.ipynb
mnist_pg.ipynb

Reference

• REINFORCE Algorithm