THIS REPO IS NO LONGER MAINTAINED. FOR A MORE STRUCTURED CODEBASE, VISIT simple-nmt, IMPLEMENTED IN PYTORCH, which is very similar to Chainer.
Implementation of recurrent neural network (RNN) and seq2seq models in Chainer. This repo is inspired by Tal Baumel's cnn (now dynet) seq2seq notebook.
The toy task is learning to reverse a string (i.e. given input "abcde", output "edcba"). Implemented models are:
- Vanilla multi-layer LSTM RNN model.
- Vanilla encoder-decoder model.
- Global-attentional encoder-decoder model (Vinyals et al.)
To run the code, please install Chainer and CuDNN first. Then evoke main.py:
$ python main.py
To run with different modes, modify main.py. Some notable variables are:
- DEVICE: the code is set to run on CPU (DEVICE = -1), set DEVICE = 0 to run on single GPU.
- LARGE: size of the data set.
- ATTEND: whether to use attention or not.
If there are any problems, email me at nguyenxuankhanhm@gmail.com