search

TuringLang / Bijectors.jl

Implementation of normalising flows and constrained random variable transformations
https://turinglang.org/Bijectors.jl/
MIT License
200 stars 33 forks source link

CompatHelper: bump compat for "Functors" to "0.2" for package test #166

Closed github-actions[bot] closed 3 years ago

github-actions[bot] commented 3 years ago

This pull request changes the compat entry for the Functors package from 0.1 to 0.1, 0.2 for package test.

This keeps the compat entries for earlier versions.

Note: I have not tested your package with this new compat entry. It is your responsibility to make sure that your package tests pass before you merge this pull request.

devmotion commented 3 years ago

Functors 0.2 allows to specify only a subset of numerical fields, maybe the implementation could be updated.

yebai commented 3 years ago

Functors 0.2 allows to specify only a subset of numerical fields, maybe the implementation could be updated.

@torfjelde Can you take a look at this, please?

devmotion commented 3 years ago

Test failures are the same as in DistributionsAD and probably caused by Distributions 0.24.11.

torfjelde commented 3 years ago

Functors 0.2 allows to specify only a subset of numerical fields, maybe the implementation could be updated.

I'm not sure what's being referred to here though. Are there any bijectors where this is the case?

devmotion commented 3 years ago

I'm not sure what's being referred to here though. Are there any bijectors where this is the case?

@torfjelde I am not sure. But in any case this could be changed later, I guess? I think it makes sense to focus on adding compatibility with Functors 0.2 first since apparently Flux 0.12 does not support Functors 0.1.

torfjelde commented 3 years ago

Down to merging as soon as tests pass :+1:

devmotion commented 3 years ago

I don't know why but both on master and here Tracker tests on Julia 1.3 take forever but finish in around 30min on Julia 1.3. It seems something (maybe upstream?) caused performance regressions in this particular setup?

I guess it doesn't block this PR though, does it?

torfjelde commented 3 years ago

I don't know why but both on master and here Tracker tests on Julia 1.3 take forever but finish in around 30min on Julia 1.3. It seems something (maybe upstream?) caused performance regressions in this particular setup?

I guess it doesn't block this PR though, does it?

You mean take forever to finish on Julia 1.3, but not on Julia 1.6? Or you mean here and on master it takes long to finish, but if you run on Julia 1.3 locally it finishes in 30 mins?

But yeah, shouldn't matter nor be related to this PR, so let's just get it through.

devmotion commented 3 years ago

It fails on Julia 1.3 (both on master and in this PR) but is reasonably fast on Julia 1.6.