search

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

integrate model hooks into keras #130

Closed DeNeutoy closed 7 years ago

DeNeutoy commented 7 years ago

@matt-gardner any guesses about why this fails with Theano for just saving the model? It should have literally nothing to do with the backend. I've also recreated this locally.

matt-gardner commented 7 years ago

Sadly, no, I don't have any good ideas. It's also really bizarre that it just fails in that one case. It looks like the ModelCheckpoint isn't executing correctly, because it's not saving the weights, but it's still odd.

Well, here's an idea - validation accuracy doesn't go up, so the model doesn't get saved? There could be some small difference between the numbers output by theano vs. tensorflow which leads to the ModelCheckpoint getting run in one case but not in others, with these tiny input sizes. Probably the right thing to do here is to just set save_models to false in the tests (in common/solvers.get_solver().