deep_qa
deep_qa copied to clipboard
allenai
A deep NLP library, based on Keras / tf, focused on question answering (but useful for other NLP too)
Currently, `WordAlignmentEntailment` layers do not work properly with serialization, and should thus be fixed.
A better DataIndexer that allows for the techniques described in https://arxiv.org/abs/1703.00993 would be fantastic. to summarize, they: - index every word in the set of glove vectors, no matter if...
Using byte encoding on unicode characters could be a good idea, vs a single index for each unicode characters. Allowing for different character encodings in tokenizers that return characters would...
So that if you want, you can represents words as character sequences like `[@BEGIN@, w, o, r, d, @END@]`. This is potentially helpful for various kinds of encoders (probably not...
The intent being that it should probably be a lot faster. I think Matt Peters already has one of these that we can just use. Though, with the dynamic padding...
It'd be really nice to know where there are performance bottlenecks in your model. I think tensorflow 1.0 added some stuff that would make this relatively easy to diagnose; can...
Say you train a model on SQuAD, then want to fine-tune it on SciQ. Presumably there will be words in SciQ that you have plenty of training data for, but...