rumaak/exploring_machine_translation

Exploring neural machine translation with pytorch. LSTM RNN and RNNsearch implementations.

Neural machine translation

In this project I have attempted to implement neural machine translation algorithm(s) together with bunch of other complementary stuff - data preprocessing, etc. The Machine translation.ipynb notebook puts together all the machinery and presents the work done. Before proceeding, see Prerequisities below.

History of this repo is quite big, so it is wise to clone only the last commit. In future, I am planning to delete this repo and create a fresh new one without this issue.

Prerequisities

Datasets

• WMT14

  • download https://1drv.ms/f/s!AiQ5a2cXVytTlkZD0HQn5FdGRgB2
  • move to data/WMT14

• Multi30k

  • download https://1drv.ms/f/s!AiQ5a2cXVytTllU4usk93QbPGv1s
  • move to data/Multi30k

Environment

I highly recommend setting up a conda environment before installing packages.

• navigate to root
• pip install -r requirements.txt

Also, while in environment, do this: python -m spacy download en

Trained models

• LSTM RNN

  • download https://1drv.ms/f/s!AiQ5a2cXVytTmUHOdj_FIim0ge2P
  • move to models/LSTM_RNN

• RNNsearch

  • download https://1drv.ms/f/s!AiQ5a2cXVytTmUTIIBY09Kq4NSC4
  • move to models/RNNsearch

References

• LSTM RNN model - arXiv:1409.3215v3 [cs.CL]
• RNNsearch model - arXiv:1409.0473v7 [cs.CL]