Michael Osthege

Thanks @jessegrabowski, I'd say this confirms my hypothesis: It's a bug related to broadcasted, symbolic sizes. @pymc-devs/dev-core can someone familiar with `aeppl` take a look at this? It does sound...

> Does that mean not to use pm.MutableData(....) or not to use pm.Data(...., mutable = True)? Can I use the latter with add_coords method? `pm.MutableData(....)` is just a shortcut for...

Closed by #6029

I see the benefits of a trace object with a simpler API, but the current implementation of BaseTrace/MultiTrace has some terrible incoherences. We'll always have a non-`InferenceData` data structure to...

Since `return_inferencedata=True` is now the default, I agree that the naming could be improved. Something like: * `pm.sample(return_mtrace=False)` * `pm.sample_p*_predictive(return_dict=False)` w.r.t. to the original description of the issue, I'm also...

I'm not too familiar with the 3.11.2 implementation, but when working on shape stuff for v4 I ran into problems with broadcasting of column vectors. The difference between the numpy...

@rpgoldman can you check your model again with a recent version? I would suspect that the issue has been resolved in the meantime

sounds fixed

Inside the `pre-commit.yml` it could run as its own job

The traceback above indicates Python 3.10. Are we sure that Aesara/PyMC work with 3.10? Because #5209