Porting to Keras 2 - Githubissues

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 132 forks source link

Porting to Keras 2 #260

Closed nelson-liu closed 7 years ago

nelson-liu commented 7 years ago

I did some work wrt porting deep_qa to Keras 2...

but there's still a lot of work to be done, off the top of my head:

Model loading changed, so we need to fix that
ShareableGRU was broken, so we need to fix that
Make sure the timedistributed stuff we're doing still works

This is a pretty massive PR, sorry in advance to whoever's reviewing. I tried to make my commit messages clear / have each commit message correspond to one "type" of change. Hopefully that makes it slightly easier to look at this.

nelson-liu commented 7 years ago

great, theano tests officially passing

matt-gardner commented 7 years ago

This is no longer WIP. Nelson, if you want to take a look, now's the time. Also, do you know how to exclude contrib/ from the coverage results?

nelson-liu commented 7 years ago

yeah, i don't think it's possible for me to do a thorough review here. I have faith in our unit tests, so let's merge. I'd be in favor of rerunning some of the models in example_experiments to ensure that we can still replicate the results of those papers.

nelson-liu commented 7 years ago

(feel free to press the button when you're ready)

nelson-liu commented 7 years ago

we should also probably edit the example_experiments, the biggest thing that I can think of is that output_dim is now units