Split sampling.py into sampling.py and sampling_forward.py

[michaelosthege]

Closes #6141

What is this PR about?
Splitting up the sampling.py module such that MCMC and prior/posterior predictive sampling are separated.

This also splits one of the longest-running test files.

I did not change any implementation - not even renamed a function.

For review, please check the __all__ settings, because they define what is the public API.

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 👇

Docs / Maintenance

• Split sampling.py into sampling.py and sampling_forward.py

[ricardoV94]

I would suggest sampling_forward or sampling_ancestral instead of sampling_predictive and to put draw in the same group (so just 2 modules, not 3). Predictive refers to the goal of sampling whereas forward/ancestral refers to the method of sampling.

[michaelosthege]

@ricardoV94 we need three modules because sampling and sampling_predictive share some dependencies.

Tearing the tests apart might be the bigger problem here..

I like sampling_predictive because it's more intuitive for noobs what to expect to find in there.. But feel free to push the branch - I'm going biking for a few hours now 🚴‍♀️

[ricardoV94]

The predictive label is a bit misleading though.

You can equally easily use mcmc to do predictive sampling (e.g., if you want to incorporate transforms/potentials) or use sample_*_predictive to do posterior inference (e.g., if you have a closed form posterior)

[codecov]

Codecov Report

Merging #6257 (49b826a) into main (ea721e4) will increase coverage by 0.03%.
The diff coverage is 95.76%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6257      +/-   ##
==========================================
+ Coverage   94.14%   94.17%   +0.03%     
==========================================
  Files         100      102       +2     
  Lines       21239    21472     +233     
==========================================
+ Hits        19996    20222     +226     
- Misses       1243     1250       +7     

Impacted Files	Coverage Δ
pymc/sampling_forward.py	95.55% <95.55%> (ø)
pymc/parallel_sampling.py	87.32% <100.00%> (+0.29%)	⬆️
pymc/sampling_jax.py	97.18% <100.00%> (ø)
pymc/smc/kernels.py	97.37% <100.00%> (ø)
pymc/tests/distributions/test_mixture.py	99.35% <100.00%> (+<0.01%)	⬆️
pymc/tests/distributions/test_multivariate.py	99.44% <100.00%> (ø)
pymc/tests/distributions/test_timeseries.py	99.50% <100.00%> (+<0.01%)	⬆️
pymc/variational/opvi.py	87.01% <100.00%> (-0.03%)	⬇️
pymc/tests/stats/test_convergence.py	100.00% <0.00%> (ø)
pymc/stats/convergence.py	92.92% <0.00%> (+2.02%)	⬆️

[ricardoV94]

What about moving the few items left in sampling_utils to utils and getting rid of that third file?

[michaelosthege]

What about moving the few items left in sampling_utils to utils and getting rid of that third file?

These itemw are used by both of the others - I think it's nice to keep sampling.py and sampling_forward.py fully independent.. After all these years, you know

[ricardoV94]

Self-approval, someone else please review. Commits should be squashed

[michaelosthege]

Diff looking good