ChatBot Trained using Seq2Seq Model on Mahabharatha Subtitles
- RNN Encoder-Decoder Model(Seq2Seq)
- MultiRNNCell GRU Cell
- Attention Decoder and Embedding Layer with Buckets
- Sampled SoftMax Loss and SGD Optimizer
- Beam Search Response Construction
The Data has been prepared from Mahabharatha Subtitles. All pre-processing and post-processing steps are available in prep_data.py file.
MBT SUBS: https://www.opensubtitles.org/en/ssearch/sublanguageid-all/idmovie-63130
It can be used to train on any set of subtitles within the same file.
- Download pre trained Model and extract in the checkpoints folder
- Run with 'python3 mbt_bot.py --m=chat'
- There are 3 modes, train, chat and api(for rest api using flask)