Move distributions implemented here to "distributions"?

[aloctavodia]

I am talking about this https://github.com/pymc-devs/distributions

[jessegrabowski]

Distributions is not meant to be "pymc ready" right? It would be good to have all the required functions for the distributions here, but users likely still want to have model compatible implementations

[aloctavodia]

Exactly, I was very imprecise. I mean to move "the core" of the distribution, the logpdf, logcdf, etc.

[jessegrabowski]

ah yes then i support this

[maresb]

I'm a bit confused. Are we AePPLing PyTensor again? 😂

Would the desired pattern be that from PyMC we do something like:

class Normal(Continuous):
    from distributions.normal import logp, logcdf, ...

[ricardoV94]

We're not AePPling, we are splitting the core probability expressions so they can be used elsewhere.

Aeppl was about looking at arbitrary pytensor random graphs, and deriving probability expressions. PyMC still does that. But at the end of the day you usually derive some expression in terms of a core density function, say a normal_logpdf, that bootstraps everything.

That bootstrapped thing will live in distributions, but the logic to derive stuff is still pymc specific.

PyMC = PPL
pytensor-distributions = Just a bunch of pytensor functions. No magic derivation there