Initiate v4.0.0b3 release

[fonnesbeck]

Version bump for v4.0.0b3

[codecov]

Codecov Report

Merging #5562 (784fe6f) into main (620546b) will decrease coverage by 0.87%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #5562      +/-   ##
==========================================
- Coverage   87.26%   86.39%   -0.88%     
==========================================
  Files          81       81              
  Lines       14244    14243       -1     
==========================================
- Hits        12430    12305     -125     
- Misses       1814     1938     +124     

Impacted Files	Coverage Δ
pymc/__init__.py	100.00% <100.00%> (ø)
pymc/sampling_jax.py	0.00% <0.00%> (-98.43%)	⬇️
pymc/parallel_sampling.py	86.71% <0.00%> (-1.00%)	⬇️
pymc/bart/utils.py	85.40% <0.00%> (+2.79%)	⬆️

[michaelosthege]

We should prominently mention the change of interpretation for the size kwarg with multivariate distributions.
And that VI is available again.

[ghost]

@michaelosthege are those not in the release notes, or are you just talking about the text associated with the release on the releases page?

[twiecki]

I think we forgot to add them, VI is still in the list of unported.

[fonnesbeck]

The size kwarg change is already in the release notes.

[twiecki]

Upon reconsideration, we probably don't need to mention it at all because it doesn't add new functionality, we just need re move it from the list of not-working things.

[ghost]

Probably fine to leave it under maintenance. Users likely will not notice something missing from the non-working list but will notice it here, and VI is a pretty big feature to have working again.

[ricardoV94]

Oops, I pushed a commit that removed this change. I do agree with @twiecki, this changelist should reflect the difference between current and v3, and as such it does not make sense to say something is still working...?

[ricardoV94]

Cholesky distributions also still show up in the not yet working, but they are!

[ricardoV94]

Actually... several algorithms of VI are not yet working... so maybe we should replace the old notice by something like:

### Not-yet working features
...
- Some Variational inference algorithms (see PR [#4582](https://github.com/pymc-devs/pymc/pull/4582))
...

[ghost]

Which are not working, and do we plan on making them work or are they gone forever?

[twiecki]

Gone forever, most likely.

[ricardoV94]

Normalizing flows:

pymc/pymc/variational/approximations.py

Line 363 in ef5f91b

raise NotImplementedInference("Normalizing flows are not yet ported to v4")

[ricardoV94]

And these:

pymc/pymc/variational/opvi.py

Lines 959 to 962 in e987950

	if self.local:
	raise NotImplementedInference("Local inferene aka AEVB is not supported in v4")
	if self.batched:
	raise NotImplementedInference("Batched inferene is not supported in v4")

[ricardoV94]

And, finally, this one, which seems minor:

pymc/pymc/variational/approximations.py

Lines 274 to 277 in e987950

	if deterministic:
	raise NotImplementedInference(
	"Deterministic sampling from a Histogram is broken in v4"
	)

[ricardoV94]

My bad:

Suggested change

[ghost]

OK, should we leave these with NotImplemented exceptions? If Thomas is correct in anticipating they won't be revived any time soon, perhaps they should be removed from the code base, given that this is a major release.

[twiecki]

I think we should remove it, yes.

[twiecki]

But we can merge this.