REGR: resample apply fails with KeyError/AttributeError for function that works on DataFrame but fails on Series

[jorisvandenbossche]

A resample apply accessing a column of the subsetted dataframe fails (bubbling up the error from first trying the function on a single column, i.e. a Series):

In [25]: df = pd.DataFrame({"col": range(10)}, index=pd.date_range("2012-01-01", periods=10, freq="20min"))  

In [26]: df.resample("H").apply(lambda group: len(group['col'].unique()))   
...
KeyError: 'col'

In [27]: df.resample("H").apply(lambda group: len(group.col.unique())) 
...
AttributeError: 'Series' object has no attribute 'col'

This seems a regression, or was there a deliberate change in apply to only apply it on the columns and no longer on the full DataFrame ?
cc @jbrockmendel

[jbrockmendel]

i dont think this was intentional

[simonjayhawkins]

The PR to fix this issues fails on the backport. #37305

since this regression is from 0.25.3 to 1.0.0, we could decide not to backport.

the issue is the index frequency is None on 1.1.x here...

pandas/pandas/core/resample.py

Line 375 in 055651b

result = grouped.apply(how, *args, **kwargs)

@jbrockmendel any ideas where changes in master could have corrected this?

[simonjayhawkins]

Note: the regression reported here is probably due to #29077

[simonjayhawkins]

maybe #35563 is related but that still fails on master