PERF/REF: groupby sample

[mzeitlin11]

• closes REF: Move sampling logic into algorithms.py #34483
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

Benchmarks:

       before           after         ratio
     [d61ace50]       [7fb839ce]
     <master~4>       <gb_sample>
-         116±5ms         29.8±3ms     0.26  groupby.Sample.time_sample
-        808±20ms         75.9±1ms     0.09  groupby.Sample.time_sample_weights

[mzeitlin11]

This was pretty much moved as is, with the only change being to convert to an ndarray earlier for better performance on later validation steps

[jbrockmendel]

im really not a fan of Series/DataFrame methods living in this file. are there any other natural homes for this?

[mzeitlin11]

util/_validators is almost a good fit, but while this validates, it also returns a modified input. Could go in core/common? But still learning where everything lives, happy to move if anyone has a better location

[jbrockmendel]

looks like this function is the only one that really depends on Series/DataFrame; could it just stay inside the NDFrame method? the others could go in e.g. core.array_algos.sample

[mzeitlin11]

That would be nicer, but the issue is that the weights processing also needs to be called from groupby sample as well.

[jbrockmendel]

Another option would be to implement something like a core.methods directory for Series/DataFrame methods that have been refactored to their own files (e.g. describe). I think algos.SelectN might make sense in something like that

[mzeitlin11]

I like that idea - have moved to core/sample.py (on same level as describe) for now. If others like this organization, I can follow up moving describe and sample (and maybe others like you mention) to core/methods.

[mzeitlin11]

Conditionals here were refactored for mypy (but IMO clearer to follow as well)

[jreback]

very nice @mzeitlin11