PERF: groupby.transform broadcast perf

[lukemanley]

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

groupby.transform broadcasts user-defined functions by passing a repeated array to np.concatenate. Passing the array once to np.tile and letting numpy repeat the array is quite a bit faster.

import numpy as np
import pandas as pd

N = 1_000_000

df = pd.DataFrame(
    data=np.random.rand(N, 3),
    index=np.random.randint(0, 5, N)
)

%timeit df.groupby(level=0).transform(lambda x: np.max(x, axis=0))

302 ms ± 8.18 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)     <- main
118 ms ± 2.02 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)   <- PR

[phofl]

Nice. Could you check if we have an asv for that case?

[lukemanley]

@phofl - there are not many groupby transform asvs that hit this code path. I just added one for this case.

[phofl]

thx

[rhshadrach]

We like to target ASVs to be around 80ms, is that right? Just want to make sure, looks appropriate as-is.

[jreback]

We like to target ASVs to be around 80ms, is that right? Just want to make sure, looks appropriate as-is.

yeah generally < 100ms is good, but we have a lot that are not so can cleanup separately.

[jreback]

thanks @lukemanley