test_take

[honno]

Somewhat comprehensive test_take

[asmeurer]

Have you tried this on different libraries? I know pytorch is missing the axis argument, but numpy, cupy, numpy.array_api (git), and cupy.array_api (git) all have take that should be fully compliant.

[honno]

Have you tried this on different libraries? I know pytorch is missing the axis argument, but numpy, cupy, numpy.array_api (git), and cupy.array_api (git) all have take that should be fully compliant.

I should of mentioned I really just developed this test using top-level numpy. Just tried this on cupy, array_api_compat.numpy and array_api_compat.cupy, which all pass this test in its current iteration. Will look at dev NumPy and CuPy later—didn't realise other folks been working on the .array_api namespaces, nice!

[asmeurer]

Yeah I didn't either, though maybe we should try to nudge people to the compat library.

If this works with those libraries I'm +1 to merge.

[honno]

The .array_api namespaces are kinda weird to test with as they don't implement complex dtypes, so I'll ignore 'em for now and get this merged given it works with NumPy-proper.