PERF: statically define classes for is_dtype checks

[TomAugspurger]

This moves the definition of the functions passed to _is_dtype_type from dynamically generated functions is is_integer_dtype to top-level functions. I can't run asv right now (#33315 (comment)) but here are some timeits

master

In [2]: t = np.dtype('int64')

In [3]: %timeit pd.api.types.is_integer_dtype(t)
2.22 µs ± 51.6 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

This PR

In [3]: %timeit pd.api.types.is_integer_dtype(t)
1.54 µs ± 39.6 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

[jreback]

better as a dict i would think

[jreback]

in fact could simply instantiate classes

[jbrockmendel]

Nice, I didnt expect classes to account for that much overhead.

[WillAyd]

I can't run asv right now (#33315 (comment)) but here are some timeits

Hmm don't think that issue is the root of any ASV issues - are you getting any other error messages from the run? Maybe worth opening a dedicated issue

[jreback]

I would move the logic into the is_* functions themselves

[TomAugspurger]

Which logic would you move? It seems like _is_dtype_or_type requires a callable, and I want to keep the definition of that callable out of e.g. is_bool_dtype.

[jreback]

example instead of creating the module level variable I would simply hardcode:
lambda tipo: issubclass(tipo, np.timedelta64) right here. its much more obvious.

[jbrockmendel]

@TomAugspurger is this still something you want to merge, or is it superceded by #33400?

[TomAugspurger]

Probably a non-issue (or less pressing) after #33400 is done. I'll re-run benchmarks after that's merged.

…

On Thu, Apr 16, 2020 at 12:23 PM jbrockmendel ***@***.***> wrote: @TomAugspurger <https://github.com/TomAugspurger> is this still something you want to merge, or is it superceded by #33400 <#33400>? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#33364 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAKAOISFT56BCX3OG7EYTJ3RM45J5ANCNFSM4MDBB6BQ> .

[jbrockmendel]

needs rebase

[jreback]

ok with this if we can update to comments.

[jreback]

conceptually ok with this if you can update to comments

[jreback]

closing as stale