`clone_replace` imports yield `ImportError`?

[larryshamalama]

Description

In light of pymc-devs/pytensor#66, I believe that the PyMC codebase still has clone_replace being imported from the wrong PyTensor file. Below is an example

pymc/pymc/logprob/transforms.py

Line 45 in f96594b

from pytensor.graph.basic import Apply, Node, Variable, clone_replace

[ricardoV94]

We always pin a specific version of Aesara/PyTensor because the API is changing quickly. This will be addressed when we next update the dependency.

There's a couple of other things that need to be updated

[larryshamalama]

We always pin a specific version of Aesara/PyTensor because the API is changing quickly. This will be addressed when we next update the dependency.

There's a couple of other things that need to be updated

Okay, I see. I didn't follow PyTensor updates too closely lately. I have the habit of importing the most up-to-date PyMC and PyTensor libraries, whether that's the most efficient way or not.

Is it worth addressing this issue alone, or should it be grouped with the other updates (in light of API changes)?

[ricardoV94]

It's worth releasing again and updating what is needed in PyMC. You can't fix this issue alone because our current PyTensor dependency is incompatible with the latest version on main of PyTensor.

[larryshamalama]

Oh, right, I understand. I had not thought of that...

[larryshamalama]

This error arose when trying to create a conda environment on a Mac as a ResolvePackageNotFound was obtained...

Adding pytensor==2.8.10 under the pip section solves the problem for my Mac and the environment-dev.yml does not yield errors on my Ubuntu machine. I'm not as familiar with managing conda environments, so I leave this issue closed for now although there may be other installation challenges related to this issue