PERF: Rolling._apply

[jbrockmendel]

• closes PERF: Rolling._apply iterates over blocks _and_ columns #42736
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

from asv_bench.benchmarks.rolling import *

self = Engine()
self.setup("DataFrame", "float", sum, "cython", "sum")
self.time_rolling_methods("DataFrame", "float", sum, "cython", "sum")

%timeit self.time_rolling_methods("DataFrame", "float", sum, "cython", "sum")
150 µs ± 889 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)  # <- PR
224 µs ± 3.18 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)  # <- master

cc @mroeschke

[mroeschke]

There's an _apply_series which looks like a 1D case of _apply_blockwise; can that be refactored into _apply_blockwise?

_apply_blockwise maybe should be renamed to _apply_columnwise

[jbrockmendel]

Updated timings:

from asv_bench.benchmarks.rolling import *

self = Engine()
self.setup("DataFrame", "float", sum, "cython", "sum")
self.time_rolling_methods("DataFrame", "float", sum, "cython", "sum")

%timeit self.time_rolling_methods("DataFrame", "float", sum, "cython", "sum")
209 µs ± 8.64 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)  # <- master
147 µs ± 3.32 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)  # <- PR, consolidate=True
133 µs ± 1.95 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)  # <- PR, consolidate=False

I'm OK with consolidating at that cost.

[mroeschke]

LGTM. Does this deserve a whatsnew note?

[jbrockmendel]

Does this deserve a whatsnew note?

either way i guess, i can add something in the next "assorted follow-ups" PR

[mroeschke]

Sure that sounds good. Thanks @jbrockmendel