Stack seq2seq encoders to give sentence selector more depth

allenai / deep_qa

A deep NLP library, based on Keras / tf, focused on question answering (but useful for other NLP too)

Apache License 2.0

404 stars 133 forks source link

Stack seq2seq encoders to give sentence selector more depth #277

Closed nelson-liu closed 7 years ago

nelson-liu commented 7 years ago

Matt suggested to stack some seq2seq encoders to give the model more depth; this does so in a similar manner to the BiDAF code.

There's a funky error going on right now wrt masking the 4D input to the seq2seq encoder, and I can't quite figure it out. It complains that the mask is 2d, and the input itself is 4d...perhaps we need a EncoderWrapper specialized for seq2seq_encoders?

matt-gardner commented 7 years ago

Hmm, I guess pyhocon is ok with python casing on booleans. I'd probably stick with valid json where possible, though.

nelson-liu commented 7 years ago

I'm happy with this --- want to merge, @matt-gardner ?