Updated batched seq2seq example
This is my version of pytorch seq2seq tutorial.
New features:
- validation data. Separate dataset for validation is used, this gives much more accurate picture of model training than just using training data
- refactored code:
- using dataloaders
- separate classes for data management
- learner class for model creation and training
- data tokenization is multicored
- possibility to use pretrained wordvectors
- added teacher forcing
For more details, explanations see my article.
See example in example_seq2seq.ipynb. Example data is from IMDB dataset.
Inspiration and some of the code is from fastai and from the original tutorial.