ENH: Add table-wise numba rolling to other agg funcions

[mroeschke]

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

Timings for a wide table

import pandas as pd
import numpy as np

df = pd.DataFrame(np.random.rand(10, 10**5))

roll_single = df.rolling(2, method="single")
roll_table = df.rolling(2, method="table")

%timeit roll_single.mean()

4.92 s ± 463 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

roll_single.mean(engine="numba", engine_kwargs={"nopython": True, "nogil": True, "parallel": True})

%timeit roll_single.mean(engine="numba", engine_kwargs={"nopython": True, "nogil": True, "parallel": True})

5.72 s ± 430 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

roll_table.mean(engine="numba", engine_kwargs={"nopython": True, "nogil": True, "parallel": True})

%timeit roll_table.mean(engine="numba", engine_kwargs={"nopython": True, "nogil": True, "parallel": True})

10.3 ms ± 1.23 ms per loop (mean ± std. dev. of 7 runs, 100 loops each)	

[jreback]

can you update the benchmark in the top of the PR (e.g. show single/table for cython/numba)

[jreback]

lgtm. test are failing :-< ping on green

[mroeschke]

All green @jreback. Needed to mark some of the numba tests as slow.

[8549]

Sorry if this is not the appropriate place for asking this, but will this be usable with .aggregate? Like df.rolling('1y', on='Timestamp').aggregate([np.mean, np.std, np.sum], engine='numba', engine_kwargs={'nopython': True, 'parallel': True}). Thanks in advance.

[mroeschke]

@8549 unfortunately no. PR's to support aggregate would be appreciated!