DEPR: allowing unknowns in take

[jbrockmendel]

• closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

[mroeschke]

What's the motivation for this change?

[jbrockmendel]

is_array_like is an antipattern #52834

[mroeschke]

Seems pretty heavy to deprecate this usage just to avoid is_array_like, especially since numpy.take will accept array-likes. Why can't we just replace not is_array_like with not isinstance(..., (np.ndarray, ...)) and accept we may be calling asarray on more items?

[jbrockmendel]

Why can't we just replace not is_array_like with not isinstance(..., (np.ndarray, ...)) and accept we may be calling asarray on more items?

That's a reasonable intuition, but np.asarray can have weird behaviors which we generally work around (but unfortunately have several slightly-different implementations of). My thought is to avoid that by just having the user do that casting.

[rhshadrach]

But we'll still allow e.g. a list in DataFrame.take and others? This seems inconsistent.

[jbrockmendel]

But we'll still allow e.g. a list in DataFrame.take and others? This seems inconsistent.

This is for the first arg in take, not the indices.

[rhshadrach]

I personally wouldn't expect pandas to implement taking from a general sequence (e.g. a list); I'm okay with deprecating from that perspective.

[jbrockmendel]

@mroeschke gentle ping on this and #52986

[mroeschke]

Thanks @jbrockmendel