Closed BeckySharp closed 7 years ago
the model is also likely not in the docs, I can add that tomorrow -- but you wanted it submitted :(
Home now, but i was thinking that i'll remove the script from this PR tomorrow. That said, you can review the rest when you get the chance. Thanks!
OK -- so there are a few files here which are included with other PRs (so don't feel the need to do a style-based code review on ones you've already done that for. When the other PRs pass, I'll merge upstream into this and resubmit). They're included here because I've been trying to make sure things are actually doing what I want them to do -- I get that that's ugly and can redo this, but you wanted it PR'd tonight so you could look at it. So, sorry!
I think the issue is that the TimeDistributed isn't building the inner layer (or it could be the inner layer (TupleMatch) not playing well with the TimeDistributed building, but the model only has 2 trainable weights, whereas before the newer versions it had 27. The tuple_match test is passing ok -- feel free to check that though to make sure that it's correct and sufficient.
thanks!