ENH: Having rolling accepts pd.IntervalIndex on a sorted datetime column

[yohplala]

Describe the solution you'd like

It is currently possible to define custom window rolling as explained in documentation.

It is required to subclass BaseIndexer and define ones' own get_window_bounds.

In case windows' bounds are defined by timestamps, pandas already provide a nice API to define them as pd.IntervalIndex.

Requested feature is to have rolling accepting pd.IntervalIndex in window parameter, that would be used in combination with on parameter to define the datetime-like column (already the case apparently)

API breaking implications

window parameter accepts pd.IntervalIndex

Additional context

This operation could be managed by use of pd.IntervalIndex and pd.cut but this combination does not take advantage that:

• intervals are sorted and non overlapping
• datetime-like column is sorted
  which allows then to go through the DataFrame only once, managing each intervals one after the other.

I think that rolling naturally does.

[yohplala]

Hi,
Sorry, I think I made a mix between 'rolling' and 'resample/groubly'. I have open ticket #41212 with a similar request, but focused on 'resample' and providing a more detailed use case. I am closing this one. Sorry for the mistake.