Upgrade to aesara=2.8.2 and aeppl=0.0.35

[Armavica]

What is this PR about?
Adapting to the new aesara API, fixes #6053.

Checklist

• Explain important implementation details 👆
• Make sure that the pre-commit linting/style checks pass.
• Link relevant issues (preferably in nice commit messages)
• Are the changes covered by tests and docstrings?
• Fill out the short summary sections 👇

Major / Breaking Changes

• ...

Bugfixes / New features

• ...

Docs / Maintenance

• Upgrade to aesara=2.8.2 and aeppl=0.0.35

[twiecki]

Thanks for taking this on @Armavica!

[codecov]

Codecov Report

Merging #6059 (a0bb1db) into main (f7315a3) will increase coverage by 0.50%.
The diff coverage is 98.07%.

@@            Coverage Diff             @@
##             main    #6059      +/-   ##
==========================================
+ Coverage   89.06%   89.56%   +0.50%     
==========================================
  Files          72       72              
  Lines       12931    12941      +10     
==========================================
+ Hits        11517    11591      +74     
+ Misses       1414     1350      -64     

Impacted Files	Coverage Δ
pymc/distributions/dist_math.py	87.42% <80.00%> (ø)
pymc/aesaraf.py	92.19% <100.00%> (ø)
pymc/distributions/continuous.py	97.50% <100.00%> (+0.02%)	⬆️
pymc/distributions/logprob.py	97.72% <100.00%> (ø)
pymc/distributions/multivariate.py	92.00% <100.00%> (ø)
pymc/distributions/simulator.py	87.50% <100.00%> (ø)
pymc/distributions/timeseries.py	78.64% <100.00%> (ø)
pymc/gp/cov.py	98.09% <100.00%> (ø)
pymc/gp/util.py	96.55% <100.00%> (ø)
pymc/math.py	70.04% <100.00%> (ø)
... and 6 more

[Armavica]

Now the only tests that do not pass are the ones related to the random variable WeibullRV of aesara which might need fixing: aesara-devs/aesara#1141

[Armavica]

So apparently, as discussed in the linked aesara issue, the WeibullRV has been reparameterized to only keep the shape parameter: the scale parameter disappeared.

So I am not sure what to do. The two options that I see are:

• Keep the current PyMC parameterization and reparameterize for aesara on the fly. Pros: no breaking change and keep a consistent scale parameter among most continuous distributions. Cons: It might be rather hackish code-wise.
• Propagate the change of parameterization to PyMC, to make Weibull require only one parameter. Pros: one-to-one mapping with aesara. Cons: breaking change, inconsistencies between distributions with a scale parameters and those without.

What are your thoughts?

[ricardoV94]

We already have a weibull with scale:

pymc/pymc/distributions/continuous.py

Line 2521 in f7315a3

class WeibullBetaRV(WeibullRV):

It seems we only need to implement the logp method (you can copy the one that was being used in aeppl until recently)

[Armavica]

Thank you for your help. The tests are now all passing, so I think that the PR is ready for review.

It took me some time to figure out what I had to do for this Weibull distribution, and I still don't understand how everything works… So I hope that I did what you meant, but otherwise I would be happy to redo it differently.

[Armavica]

I realized that I forgot to address some deprecations which only raised warnings and not hard errors. So I am going to pass this PR back into draft mode until this is done.

[ricardoV94]

Awesome work!

[twiecki]

Very cool, and already in our new release: https://github.com/pymc-devs/pymc/releases/tag/v4.1.6