Include Op in message when raising NotImplementedError from grad

[ricardoV94]

Description

When grad fails because a specific Op doesn't support it it's quite difficult to figure out which one it was without jumping on an interactive debugger

https://discourse.pymc.io/t/avoiding-non-nuts-samplers-when-broadcasting-a-multiplication-operation/14245/4?u=ricardov94

We should return something like NotImplementedError(f"grad not implemented for Op {self}") in a base class.

Or use the grad_not_implemented functionality that exists for partial grad implementations which also provides a more readable error IIRC

[tanish1729]

hello! i'll take up this issue. i understand the issue and also saw the main issue on pymc. how should i start solving this?

[ricardoV94]

You should tweak this base case NotImplementedError, to include self in the message:

pytensor/pytensor/graph/op.py

Line 360 in 090b844

raise NotImplementedError()

[tanish1729]

Im trying to use the grad_not_implemented function but im slightly confused about the implementation.
what is the inputs parameter to the grad function so that i can use it while passing to the grad_not_implemented as used in some other places.

[tanish1729]

i made some commits to my fork but they arent showing up in this issue for u to review? 😕

[ricardoV94]

You need to open a pull request from your fork: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/creating-a-pull-request-from-a-fork