[BUG] Imputer imputes too many pd-multiindex missing values about sktime HOT 4 CLOSED

Ah, indeed! So you'd need to group it again before the bfill, right?

Would you like to open a PR with a fix? Seems like the correct diagnosis already!

Your example would also make a good test.

from sktime.

Ah, indeed! So you'd need to group it again before the bfill, right?

Would you like to open a PR with a fix? Seems like the correct diagnosis already!

Exactly! I'm happy to do so, I'll pick it up tomorrow.

from sktime.

Excellent.

Regarding the doc/logic mismatch, the standard way to address this is to change the doc, not the logic - to avoid to break currently working code.

Changing the logic can be done later, but should be accompanied by warnings etc - if we decide to do that.

from sktime.

EDIT: some profiling shows that the current solution is the fastest, even for a small dataframe. df.index.levels > 4 is approximately 1/3 slower than isinstance(df, MultiIndex):

>>> df = pd.DataFrame(dict(x=[1,2,3], y=[4,5,6]))
>>> timeit(lambda: isinstance(df, pd.MultiIndex), number=100000000)
6.388806922000015
>>> timeit(lambda: df.index.nlevels > 1, number=100000000)
9.203695821999986

Olde comment:
As a side-note, I'm not wasn't sure whether this is the best way to check for presence of a MultiIndex.

from sktime.