PERF: Do not init cache in RangeIndex.take

[GianlucaFicarelli]

Improve performance when passing an array to RangeIndex.take, DataFrame.loc, or DataFrame.iloc and the DataFrame is using a RangeIndex

• closes PERF: RangeIndex cache is written when calling df.loc[array] #53387
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

Performance test:

import numpy as np
import pandas as pd

print(f"index creation")
%timeit pd.RangeIndex(100_000_000)

indices = [90_000_000]
print(f"{len(indices)=}, index creation + first execution")
%timeit pd.RangeIndex(100_000_000).take([90_000_000])

print()
ind = pd.RangeIndex(100_000_000)
indices = np.array([90_000_000])
print(f"{len(indices)=}, first execution")
%time ind.take(indices)
print(f"{len(indices)=}, other executions")
%timeit ind.take(indices)

print()
ind = pd.RangeIndex(100_000_000)
rng = np.random.default_rng(0)
indices = rng.integers(0, 100_000_000, 1_000_000)
print(f"{len(indices)=}, first execution")
%time ind.take(indices)
print(f"{len(indices)=}, other executions")
%timeit ind.take(indices)

Output (pandas 2.0.1):

index creation
3.58 µs ± 9.11 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)
len(indices)=1, index creation + first execution
204 ms ± 41.2 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

len(indices)=1, first execution
CPU times: user 54.9 ms, sys: 147 ms, total: 202 ms
Wall time: 202 ms
len(indices)=1, other executions
3.79 µs ± 21.6 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

len(indices)=1000000, first execution
CPU times: user 80.4 ms, sys: 155 ms, total: 236 ms
Wall time: 236 ms
len(indices)=1000000, other executions
28.2 ms ± 193 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

Output (this PR):

index creation
3.54 µs ± 15.7 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)
len(indices)=1, index creation + first execution
14.6 µs ± 53.9 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

len(indices)=1, first execution
CPU times: user 38 µs, sys: 0 ns, total: 38 µs
Wall time: 41.5 µs
len(indices)=1, other executions
7.96 µs ± 16.7 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

len(indices)=1000000, first execution
CPU times: user 2.92 ms, sys: 1.99 ms, total: 4.91 ms
Wall time: 4.93 ms
len(indices)=1000000, other executions
2.17 ms ± 277 ns per loop (mean ± std. dev. of 7 runs, 100 loops each)

[github-actions]

This pull request is stale because it has been open for thirty days with no activity. Please update and respond to this comment if you're still interested in working on this.

[jbrockmendel]

Couple questions, otherwise LGTM

[GianlucaFicarelli]

Couple questions, otherwise LGTM

Thanks, I've answered all the pending questions.

[jbrockmendel]

thanks @GianlucaFicarelli