ENH: Add first and last aggregations to Rolling and Expanding

[snitish]

• closes Feature request: first & last for rolling() #33155 (Replace xxxx with the GitHub issue number)
• 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/v3.0.0.rst file if fixing a bug or adding a new feature.

[mroeschke]

Thanks for starting this @snitish.

Curious if we even need a Cython function for this? It seems we can replicate this with a take on the original data accounting for min_periods/window size

[snitish]

@mroeschke would take be able to handle NAs efficiently though? i.e. perform the aggregation in O(N) time as opposed to O(NW) (W = window)

[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.

[snitish]

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.

bumping this for visibility.

[snitish]

Some quick benchmarks I ran:

In [5]: ser = pd.Series(np.random.randn(10000))

In [6]: %timeit ser.rolling(5).apply(lambda x: x[~x.isna()].iloc[0]) # using custom lambda
831 ms ± 4.21 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [7]: %timeit ser.rolling(5).first()
150 μs ± 881 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

In [8]: %timeit ser.rolling(5).last()
144 μs ± 1.14 μs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

[mroeschke]

Sorry for the delay @snitish. Thanks for sticking with this

[snitish]

Sorry for the delay @snitish. Thanks for sticking with this

No problem at all. Thanks for approving!