Kyle Daruwalla

I agree that the weights are the main blocker. I can make time for that, but I think merging this PR would be a strict improvement in the amount of...

Getting the same issue with `Homebrew.jl#master` ``` julia> versioninfo() Julia Version 1.1.0 Commit 80516ca202 (2019-01-21 21:24 UTC) Platform Info: OS: macOS (x86_64-apple-darwin14.5.0) CPU: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz WORD_SIZE:...

Yeah I just ran into the exact same issue today. Julia 1.6.1 and Zygote 0.6.12.

Unlike the original discourse, I'm not defining my own `pmap` or its gradient. For me it is just erroring using the released Zygote with the same message as the discourse...

Why not just do `@ref "Performance tips"` which is the Documenter default?

That's not the intended use for `Flux.train!`. This function is meant to iterate over an entire epoch, not a single batch. Try writing your loop as ```julia function train_loop(model, optimizer,...

Can you delete the `test/dropout.jl` file that you have added? So far, this PR is on the correct trend, but not targeting the right set of tests to replace. Specifically,...

That's because that test is testing for the wrong thing. The scaling parameters in the `AlphaDropout` layer rescale the output so that `mean(y) = 0`. This is exactly what is...

First, `evalwgrad(m, x)` and `m(x)` produce the same output. The first just computes that output under a gradient context. How they relate to each other, which is what `mean(y) =...

An `Activation` layer won't help if it wraps an anonymous function. It's a wrapper so it just pushes the issue one node deeper in the tree. This kind of solution...