DOC: Summing by level on a category index produce extra columns.

[gepcel]

Generate a dataframe:

import pandas as pd
import numpy as np
column_cat = pd.CategoricalDtype(['One', 'Two', 'Three', 'Four'], ordered=True)
groups = list('AABBB')
numbers = pd.Categorical(['One', 'Two','One', 'Two', 'Three'], 
                         categories=column_cat.categories, ordered=True)
d = pd.DataFrame(data=np.arange(15).reshape(3, 5), 
                 columns=pd.MultiIndex.from_arrays([groups, numbers]))

	A		B
	One	Two	One	Two	Three
0	0	1	2	3	4
1	5	6	7	8	9
2	10	11	12	13	14

Now do d.sum(axis=1, level=1), get the following result:

	One	Two	Three	Four
0	2	4	4	0
1	12	14	9	0
2	22	24	14	0

Which produces a non existing column Four.

pandas 1.1.1, pandas 1.1.2

[phofl]

Is this done on purpose? Looks related to the explanation in https://pandas.pydata.org/pandas-docs/stable/user_guide/categorical.html#operations

[MarcoGorelli]

I think so, but the user guide says

Series methods like Series.value_counts() will use all categories, even if some categories are not present in the data [...] Groupby will also show “unused” categories:

yet here we have a DataFrame method. Perhaps a little example like this one would be a good addition there

[phofl]

Then this is not a bug, isn't it?