Pytorch implementation of Seq2Seq model for conversational agents
CPU or CUDA version 8.0, Python version 3.6
Unzip data/cornell_movie_dialogs_corpus.zip.
Run python src/data.py for processing the data.
Run python src/runner.py for training.
Print conversation examples every epoch (make sure it's working)Print BLEU scores every epochChange SGD to AdamChange padding to maskingAdd memory padAdd attention mechanismOptimize memory usage- Resolve out of memory issue
- Implement baseline model (Seq2Seq with attention and beam search)
- Use settings from this repo
- Handle/Remove very short sequences (length <= 3) in training/prediction
- Add TensorBoard