Prior: Differentiate dims=() and dims=None

[ricardoV94]

This is necessary so we can provide a safe default for the user. Also xdist variables can infer dims implied by the parameters just fine, no reason to require redundancy.

This is a major change as code elsewhere may assume dims are never None

[ricardoV94]

Unrelated failure due to pymc-devs/pymc#8106

[williambdean]

Is this a breaking change for configurations that currently use Prior class in deployment?

[williambdean]

What is the difference between Prior(...).to_dict() and Prior(..., dims=()).to_dict()

[ricardoV94]

Is this a breaking change for configurations that currently use Prior class in deployment?

If dims were not specified, potentially yes, depends what the code on the receiving end of Prior does. If they were explicit, nothing changed

[ricardoV94]

What is the difference between Prior(...).to_dict() and Prior(..., dims=()).to_dict()

I'll check and test. Prior.from_dict(Prior().to_dict()) should be a perfect inverse

[ricardoV94]

What is the difference between Prior(...).to_dict() and Prior(..., dims=()).to_dict()

Pushing new changes, the former doesn't include dims in the dict, the latter includes "dims": ()

[ricardoV94]

@williambdean this will break code if and only if it tries to access dims directly and doesn't expect None as a possibility.

Otherwise cases where dims were not specified but were supposed to be () will behave as before. The only thing that is different is that code that would fail before, because dims=() was invalid (e.g., parameters imply more dims) will work (in some contexts) or only fail later.

I can't think of a cleaner transition strategy

[williambdean]

LGTM!