[ENH] (WIP) Creating a new Bayesian Regressor with PyMC as a backend

[meraldoantonio]

Reference Issues/PRs

#7

What does this implement/fix? Explain your changes.

This WIP PR implements a Bayesian Linear Regressor with PyMC as a backend

Does your contribution introduce a new dependency? If yes, which one?

Yes - it depends on PyMC family: PyMC itself, XArray and ArviZ

What should a reviewer concentrate their feedback on?

The design of the BayesianLinearRegressor. Especially:

1. The introduction of the priors. For now, the class hardcodes the priors. We need to think about the way in which the users should inject their own priors.

Did you add any tests for the change?

Not yet

Any other comments?

N/A

PR checklist

For all contributions

• I've added myself to the list of contributors with any new badges I've earned :-)
  How to: add yourself to the all-contributors file in the skpro root directory (not the CONTRIBUTORS.md). Common badges: code - fixing a bug, or adding code logic. doc - writing or improving documentation or docstrings. bug - reporting or diagnosing a bug (get this plus code if you also fixed the bug in the PR).maintenance - CI, test framework, release.
  See here for full badge reference
• [ X] The PR title starts with either [ENH], [MNT], [DOC], or [BUG]. [BUG] - bugfix, [MNT] - CI, test framework, [ENH] - adding or improving code, [DOC] - writing or improving documentation or docstrings.

For new estimators

(This is not yet done)

• I've added the estimator to the API reference - in docs/source/api_reference/taskname.rst, follow the pattern.
• I've added one or more illustrative usage examples to the docstring, in a pydocstyle compliant Examples section.
• If the estimator relies on a soft dependency, I've set the python_dependencies tag and ensured
  dependency isolation, see the estimator dependencies guide.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[fkiraly]

would inverse gamma not be more standard here, as it is conjugate to the normal?

[fkiraly]

Nice contribution!

Some high-level points:

• could you split the notebook off into a chained PR, based on the estimator PR? The notebook may require some more review (time), and should not block the estimator
• in the estimator, I would kindly ask you to remove the visualization dependencies from python_dependencies, and instead introduce dependency checks in the methods that need them? This way, users do not need to visualisation dependencies when using the model in a deployment pipeline.

[fkiraly]

Strange import error - is this related to an upper bound of any of the imports implied by 3.9, e.g., scipy?

[meraldoantonio]

Strange import error - is this related to an upper bound of any of the imports implied by 3.9, e.g., scipy?

Apparently there is a bug with Arviz 0.17 and scipy>=1.13 (source 1) (source 2).

The bug is no longer present in Arviz 0.18 but this requires Python 3.10 and above.

As a temporary solution, I've locked the scipy version in all-extras in project.toml

[fkiraly]

Makes sense.

From a maintenance perspective, applying the version bound in the pyproject.toml is not a good solution, since the lock is implied only by a single estimator, and not by scipy itself.

Could you add the lock instead in the python_dependencies tag of the estimator, and revert the changes to pyproject?

[meraldoantonio]

Makes sense.

From a maintenance perspective, applying the version bound in the pyproject.toml is not a good solution, since the lock is implied only by a single estimator, and not by scipy itself.

Could you add the lock instead in the python_dependencies tag of the estimator, and revert the changes to pyproject?

Makes sense! But I've tried this a couple of times and for some reason, without the pyproject.toml lock, the test framework keeps installing the "wrong" version of scipy (version 1.13.1), even after specifying "scipy<=1.12.0" in the python_dependencies tag...

It might be that other libraries are pulling in a conflicting version, but I haven't managed to find the exact cause..

Any ideas?

[fkiraly]

Any ideas?

Why are you trying to bound scipy instead of arviz? I would simply bound arviz>=0.18, based on your statements, as well as python_version >= 3.10

[fkiraly]

PS: why did you close the notebook PR? That was a nice notebook, and indeed it would be nice as separate PR.