fix import error due to moving progress bar functions in latest pymc

[Dekermanjian]

import progress bar functions from pymc.progress_bar instead of pymc.util

This addresses the issue in #541

What I am not sure of though is what happens if a user wants to use the latest pymc-extras version with pymc<5.24.0?

[jessegrabowski]

Tests are failing because of the changed import -- my guess is we have to update a version pin?

[jessegrabowski]

What I am not sure of though is what happens if a user wants to use the latest pymc-extras version with pymc<5.24.0?

@maresb

[Dekermanjian]

Tests are failing because of the changed import -- my guess is we have to update a version pin?

@jessegrabowski Yes, the majority of tests are failing because of the version of PyMC used is <5.24.0 in the tests. There is one more issue with preliz also not being able to be imported and that is causing the tests in test_prior.py to not run.

[jessegrabowski]

Could you pin pymc>=5.24.0 in this PR?

[ricardoV94]

Preliz needs to change it's import from compile_pymc to compile

[ricardoV94]

arviz-devs/preliz#682

[ricardoV94]

You can put a temporary xfail on the preliz test, and import inside the relevant test only

[jessegrabowski]

preliz already fixed it and cut a release, but it's not out of conda-forge yet

arviz-devs/preliz#680

[ricardoV94]

I see, then just wait 1/2 days? If you're confident everything is fixed feel free to merge

[jessegrabowski]

@ricardoV94 I think something about the progress bar update is breaking pickling on windows.

[ricardoV94]

Maybe the custom sampler is broken? You can change the test to run with cores=1 to see the error message

[jessegrabowski]

No error raised when cores=1

[ricardoV94]

Odd can you reproduce locally?

[jessegrabowski]

Yeah. It happens on windows because it defaults to mp_ctx='spawn'. It will happen on linux too if you set this manually.

[ricardoV94]

And didn't happen before our changes? So odd

[jessegrabowski]

The error appears to be an infinite recursion when trying to unpickle the inner function in PointFunc. That was recently changed in this commit, which might be the actual source of the problems?

Edit: Reverting that commit (and the preceding one) solves the issue 🤔

What was the motivation for those changes?

[jessegrabowski]

Thank you @Dekermanjian !

[Dekermanjian]

Happy to help! I am glad this is resolved now!