ENH: Adding `pipe` operations to `pandas.api.typing.Rolling` instances

[brianjburrow]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

I wish I could use a pipe method on a rolling object for speeding up operations that require multiple vectorized (terminology?) operations on a rolling object.

Feature Description

import pandas as pd
import numpy as np
n_samples = 60*60*2 # 2 hours 
start_ts = pd.Timestamp("2022-01-01")
end_ts = start_ts + pd.Timedelta(f"{n_samples}s")
s = pd.Series(
    np.random.normal(0, 1, size=n_samples),
    index=pd.date_range(start_ts, end_ts, freq="1s", inclusive="left")
).rolling("1T").pipe(lambda r: r.std()/r.mean())

Alternative Solutions

I'm not aware of any.

Additional Context

Maybe a bit silly, but I've seen recommendation to used chained operations when possible. I tend to like the style, but I haven't figured out an easy way to do this particular operation without a performance hit.

I could be misinterpreting how pipe actually works, but I think it might be equivalent to calc(r) performance wise?

Thank you for taking a look!

[rhshadrach]

I could be misinterpreting how pipe actually works, but I think it might be equivalent to calc(r) performance wise?

Yes - that's correct.

I'm +1 here; cc @mroeschke

[mroeschke]

Sure a PR to add pipe would be welcome.

[jrmylow]

I'm happy to pick this one up, but would it rely on the caller to use only vectorised functions?

[rhshadrach]

would it rely on the caller to use only vectorised functions?

In using pipe, we pass the instance in question (in this case, Rolling) to the callable that the user passes. What they do with it from there on is their business 😄

[jrmylow]

take