odashi / chainer_examples

Example usages of Chainer for natural language processing.

Chainer example code for NLP

This repository is out of date and rough. I do not guarantee that these code works correctly.

I am developing a new NMT toolkit NMTKit and strongly recommend to use it instead of these samples to train neural translation models.

This repository contains some neural network examples for natural language processing (NLP) using Chainer framework.

Chainer Official (GitHub)

Making Local Client

Before running these scripts, making a local python client using pyenv is reccomended, like:

$ pyenv install 3.5.0
$ pyenv virtualenv 3.5.0 example
$ pyenv shell example
$ pip install chainer

Contents

• Machine Translation

  • mt_s2s_encdec.py - Using encoder-decoder style recurrent neural network
  • mt_s2s_attention.py - Using attentional neural network

• Word Segmentation (Tokenization)

  • seg_ffnn.py - Using feedforward neural network
  • seg_rnn.py - Using recurrent neural network

• Language Model

  • lm_rnn.py - Using recurrent neural network (RNNLM)

Contact

If you find an issue or have some questions, please contact Yusuke Oda:

• @odashi_t on Twitter (faster than other methods)
• yus.takara (at) gmail.com

chainer_examples's Issues

Attention Mechanism is Very Slow

Hi,
Your attention mechanism is quite slow. Since you compute the linear projections (aw and bw) each time although they do not change, the time is almost quadratic.

I have implemented a faster version of attention which does a lot of precomputation and would like to push it as soon as I am done testing.

Regards.

a float is required , : lm_rnn.py

Hi, I have some problem.
I compile the code that lm_rnn.py

a float is required...

File "lm_rnn.py", line 219, in
main()
File "lm_rnn.py", line 213, in main
save_rnnlm_model(args.model + '.%d' % (epoch + 1), args.vocab, args.embed, args.hidden, vocab, model)
File "lm_rnn.py", line 107, in save_rnnlm_model
print(dlm.join(fmt % x for x in row), file=fp)
File "lm_rnn.py", line 107, in
print(dlm.join(fmt % x for x in row), file=fp)
TypeError: a float is required