Ilhan Polat

I think Sagemath would be a better home for it, if it has to live in Python world since these maps are quite specific for a particular math-oriented audience. And...

For this, (I am feeling like I am parroting the same stuff), in my opinion, array API **must** provide all the testing tools including the *backend mocking suite* without installing...

> If we want to start accepting PyTorch arrays in SciPy, we need to verify not only that PyTorch is compatible with the standard, but that our implementations which use...

That's fine but then Array API linalg extension should not be there because there is no linalg backend that is not compiled in any shape in any other library. So...

OK let me try to give a similar example why these are all intertwined in my opinion. The reason why we want to have non-numpy arrays (at least for Python...

> The determinant is standardised as xp.linalg.det. So if scikit-learn is happy to require that alternative backends implement the linalg extension, they will use that. Rhetorically, then why are we...

Sure, will do. Ping me if I forget about it please.

No concerns. It's not a controversial statement. Given how conservative we are typically towards any addition backend or codebase, adopting a standard that is at its infancy, must have come...

I don't know if the build will succeed but at least to try, you need a fortran compiler on your system. That's what the error is saying.

I have no idea unfortunately. Not sure if we even support windows arm. It's not an architecture I am familiar with. EDIT: Some google results claim that Flang is available...