PERF/REGR: isin slowdown for masked type

[mzeitlin11]

xref #42708

• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

Fixes isin slowdowns mentioned in https://simonjayhawkins.github.io/fantastic-dollop/#regressions?sort=3&dir=desc&branch=1.3.x

Benchmarks:

-         550±7ms         495±10ms     0.90  algos.isin.IsInLongSeriesValuesDominate.time_isin('Float64', 'random')
-         514±8ms          449±6ms     0.87  algos.isin.IsInLongSeriesValuesDominate.time_isin('Float64', 'monotone')
-     2.22±0.07ms      1.68±0.05ms     0.76  algos.isin.IsInFloat64.time_isin('Float64', 'many_different_values')
-         208±9ms          147±8ms     0.71  algos.isin.IsInLongSeriesValuesDominate.time_isin('Int64', 'monotone')
-        395±20μs          279±4μs     0.71  algos.isin.IsIn.time_isin_categorical('boolean')
-        304±10μs          200±2μs     0.66  algos.isin.IsIn.time_isin_empty('boolean')
-        341±10μs         218±10μs     0.64  algos.isin.IsIn.time_isin('Int64')
-        339±10μs          212±3μs     0.63  algos.isin.IsIn.time_isin_categorical('Int64')
-        258±10μs          147±3μs     0.57  algos.isin.IsIn.time_isin_empty('Int64')
-         221±5μs          104±3μs     0.47  algos.isin.IsIn.time_isin('boolean')
-         155±5μs         63.4±1μs     0.41  algos.isin.IsIn.time_isin_mismatched_dtype('boolean')
-         166±7μs         63.4±1μs     0.38  algos.isin.IsIn.time_isin_mismatched_dtype('Int64')
-         499±6ms          176±2ms     0.35  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 1000, 'random_misses')
-         500±8ms          173±2ms     0.35  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 1000, 'monotone_misses')
-         125±1ms       33.2±0.8ms     0.27  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 5, 'random_hits')
-         437±6ms          116±1ms     0.26  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 1000, 'random_hits')
-         125±1ms       32.9±0.8ms     0.26  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 5, 'random_misses')
-       131±0.9ms         33.3±2ms     0.25  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 5, 'monotone_hits')
-        443±10ms        109±0.6ms     0.25  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 1000, 'monotone_misses')
-        417±10ms       81.0±0.6ms     0.19  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 1000, 'monotone_hits')
-         298±8ms       41.2±0.1ms     0.14  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 1000, 'monotone_hits')
-        343±10ms       41.0±0.3ms     0.12  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 1000, 'random_hits')
-         370±7ms       39.3±0.2ms     0.11  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 1000, 'random_misses')
-        371±10ms       36.6±0.3ms     0.10  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Int64', 5, 'monotone_misses')
-        350±10ms       33.8±0.5ms     0.10  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 5, 'monotone_misses')
-        351±10ms       32.7±0.7ms     0.09  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 5, 'monotone_hits')
-        347±10ms       32.0±0.8ms     0.09  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 5, 'random_misses')
-        357±10ms       31.7±0.5ms     0.09  algos.isin.IsInLongSeriesLookUpDominates.time_isin('Float64', 5, 'random_hits')

Not sure if we do this anywhere else, but based on this slowdown seems like empty/zeros_like(extensionarr) should be avoided (this call has extra overhead through dispatching, but also triggers an astype somewhere, which is the main perf difference).

Also as a sidenote seems like empty_like/zeros_like should be avoided in general in favor of zeros/empty, see https://stackoverflow.com/questions/27464039/why-the-performance-difference-between-numpy-zeros-and-numpy-zeros-like

But zeros_like -> zeros makes a much smaller difference than zeros_like(self) -> zeros_like(self._data)

[jreback]

can you change to this PR number

[mzeitlin11]

changed

[jreback]

thanks @mzeitlin11

[jreback]

@meeseeksdev backport 1.3.x

[lumberbot-app]

Something went wrong ... Please have a look at my logs.