BUG: Rolling .apply() with method='table' ignores min_periods about pandas HOT 2 CLOSED

Thanks for the report - further investigations and PRs to fix are welcome!

from pandas.

I put up a quick fix, but there's a larger issue with how min_periods is used with the indexers that choose the windows for rolling operations. Specifically, indexers have a method get_window_bounds that takes in the min_periods argument, but none of the implementations use it.

If an indexer ever used min_periods, it would probably cause the rolling operation to crash. Usually, these operations validate the output of get_window_bounds and raise if the number of windows returned doesn't match the number of rows in the data.

Since the indexers ignore min_periods, most rolling methods manually check min_periods elsewhere. In the case of this bug, min_periods was checked after applying the function, so the output correctly had two rows of NA, but incorrectly called the function on smaller windows. (I think there's one other case where min_periods isn't properly applied. I'll check and open a separate bug report if so.)

In my opinion, either indexers shouldn't take in min_periods at all, or they should respect it and only return window indexes that have at least min_periods periods.

from pandas.