SelectTransform extended to Symbols

[molet]

SelectTransform functionality is extended to Symbols making it applicable to AxisArrays.

[willtebbutt]

LGTM. Thanks for the PR @molet . Just a couple of requests:

1. Could you try running test_ADs on a SelectTransform with Symbols, just to ensure that it actually does work as intended.
2. Bump the patch version so we can make a release :)

[willtebbutt]

@molet Did the test_AD function not work for this code?

[molet]

1. Could you try running test_ADs on a SelectTransform with Symbols, just to ensure that it actually does work as intended.

Unfortunately test_ADs function creates the input matrices inside so I cannot use that directly to test SelectTransform with Symbols. Anyway, I added a bunch of tests for Zygote and ForwardDiff. ReverseDiff unfortunately breaks - I'm still investigating why.

[willtebbutt]

Oh I see. Yes, that is unfortunate -- our AD testing is in need of a re-vamp generally. Good to know that's a thing we've accidentally been assuming.

[willtebbutt]

Don't worry too much about ReverseDiff for now if it's being a pain -- I would just add a single test using @test_broken so we don't forget about it. Then I'm happy to approve.

[willtebbutt]

Will merge when tests pass.

[willtebbutt]

Damn. These tests failed because they take too long. @devmotion @theogf any thoughts on the best way to work around this? Do we just add more @info statements or something?

[devmotion]

Hmm yes we could try that. I guess AxisArrays in general might not be optimized for the use with Zygote - in my experience the package is quite rough and a bit suboptimal and slightly outdated. Maybe we could use some other custom arrays in the tests such as OffsetArrays or AxisKeys.

[willtebbutt]

Hmm I'm reluctant to change the array type being tested for the sake of making our tests run faster.

@molet could you print a space (I was thinking just print(" ")) after each of the files in the "transform" test set has run? This should deal with this issue.