Solving the task of machine translation (from English to Vietnamese) with a regular Seq2Seq network and a global attention-based, dot product Seq2Seq network. Both following the Encoder-Decoder architecture.
Preprocessed data courtesy of the Stanford NLP group.
TODO:
- [] Attention-based network: Compute exact loss for each sequence (i.e., don't compute loss for sequences with padding). Look into utilizing
torch.nn.utils.rnn.pack_padded_sequence - [] Regular network: Get on track with the attention network, files are in
.src/TODO/ - [] Training: Run for more epochs
- [] Training: Implement K-fold CV
- [] Training: Hyperparameter tuning
- [] Testing: Implement model evaluation for user inputs and test data
- [] Miscellaneous: More documentation + better organization
Attention-based network [01/01/2021]:
