Rémi Louf

Note that `at.reciprocal(x)` should be an alias for `at.pow(x, -1)` and `at.sqrt(x)` an alias for `at.pow(x, 0.5)` for the same consistency reasons. Consistency is going to prevent special-casing the code...

> This sounds like a new/different compilation interface and/or approach, because it basically foregoes the all `Function` machinery itself. We might be able to reuse the `aesara.function` entry-point and provide...

Here is an explanation of how to go about adding an implementation: https://github.com/aesara-devs/aesara/issues/1335#issuecomment-1344510169

Yes, we should so this via pre-commit. @mohitsaxenaknoldus you can start with a minimal configuration like this one: https://github.com/blackjax-devs/blackjax/blob/main/.gitlint which should be enough to get started and iterate on.

I planned to do that in the docs

`jax.numpy.arange` and `jax.numpy.reshape` only accept concrete values respectively for their `start`, `stop`, `step` and `shape` parameters. Concrete values are either constants or the output of a `Shape` operator or combination...

> @rlouf, did you ask about this approach (i.e. using rewrites for gradients) in a Discussion or comment in Gitter? I remember it was mentioned during the last Aesara meeting,...

By all means! You may want to do a project search for `def clone`, this is a part of the codebase I don't know very well.

It is rather rude to close an issue relevant to this repo only to keep it on your fork: https://github.com/pymc-devs/pytensor/issues/6

That looks awesome! We have an issue that's been open for a while to implement `einsum`, it would be great to have it in the library.