Chatbot implementations based on movie corpus dataset

Utils and modules for NLP, audio and multimodal processing using sklearn and pytorch

Training:

1. Make a venv and use pip install -r requirements.txt to install required packages.
2. Add Movie Triples data to: ./data/MovieTriples
3. Add emebddings file to: ./cache
4. Run in terminal: export PYTHONPATH=./ (in cloned root dir)
5. python experiments/hred/training/hred_movie_triples.py -epochs 80 -lr 0 .0005 -ckpt ./checkpoints/hred -emb_dim 300

You can also add more options:

-shared (to use shared weights between encoder and decoder)

-shared_emb (to use shared embedding layer for encoder and decoder)

-emb drop 0.2 (embeddings dropout)

-encembtrain (to train encoder embeddings)

-decembtrain (to train decoder embeddings)

(see argparser in experiments/hred/training/hred_movie_triples.py for more)

Running experiments:

1. python experiments/hred/training/hred_movie_triples.py -epochs 80 -lr 0 .0005 -ckpt ./checkpoints/ -emb_dim 300 -decr_tc_ratio -encembtrain -decembtrain -shared_emb

2. python experiments/hred/training/hred_movie_triples.py -epochs 80 -lr 0 .0005 -ckpt ./checkpoints/ -emb_dim 300