Brandon T. Willard

To first introduce Windows tests, we need to restrict them to only a single `matrix.include`; otherwise, we balloon the test matrix with long-running tests with too much redundant coverage. In...

The `Variable.tag` approach currently used in `aesara.sandbox.linalg.ops` isn't particularly good, because it affects `Variable` instances globally (i.e. outside of just the rewrite processes), although it does avoid adding extra nodes...

> One way I see this `Op` being implemented is by having it store an `Function` as a property (similar to how `Elemwise` does with it's atomic functions) and then...

> Built a rudimentary implementation: https://gist.github.com/kc611/72732f4305fe273c2da1620d6fb4e90c Yes, that's a good start, especially the use of the `HasInnerGraph` interface. To parse the NumPy signatures, we can use the same code NumPy...

@kc611, I just created an outline for the kind of `Op` that might work. You're welcome to take over that PR if you're interested in implementing this; otherwise, we can...

[This](https://stackoverflow.com/a/52087847/3006474) looks like a viable solution.

This is handled in the soon-to-be-merged https://github.com/aesara-devs/aesara/pull/1054.

Here's an example of why these `DimShuffle`s are bad: they prevent one from replacing terms in a graph with new terms that have different dimensions. ```python import aesara import aesara.tensor...

As @ricardoV94 mentioned in our discussions about `DimShuffle` lifting, removing these `DimShuffle`s created by `Elemwise.make_node` would help a lot. The basic idea is that we can move the job of...

@larryshamalama, if you provide more information about these problems in a new or existing [discussion](https://github.com/aesara-devs/aesara/discussions), we can help debug. > * With the help of @ricardoV94, we've tried installing Aesara...