BUG: I got empty Dataframe with index from the summation of empty Dataframe with MultiIndex

[tkaewplik]

I'm not sure that the current behavior is right, so I would like to ask.

When I have a non-empty Dataframe I can use sum(level=['Project', 'Task'']) to get the MultiIndex Dataframe as my expectation. But when I try to sum the empty Dataframe, I got Index Dataframe instead of MultiIndex Dataframe and It has no index name and column name in there.

Is this the correct behavior or this is a bug. If this is the correct one, do you have any suggestions for me to handle this?

Thank you in advance.

Code Sample

pd.__version__
# non-empty data
data = pd.DataFrame([(1,1,1,1,1,1),(2,2,2,2,2,2)],columns=('Project', 'Task', 'Increasing Length', 'Start Period', 'Period', 'Value'))
data.set_index(['Project', 'Task', 'Increasing Length', 'Start Period', 'Period'], inplace=True)
sum_data = data.sum(level=['Project', 'Task'])
print(sum_data)

# empty data
empty_data = pd.DataFrame([],columns=('Project', 'Task', 'Increasing Length', 'Start Period', 'Period', 'Value'))
empty_data.set_index(['Project', 'Task', 'Increasing Length', 'Start Period', 'Period'], inplace=True)
# sum of empty data
sum_empty_data = empty_data.sum(level=['Project', 'Task'])
print(sum_empty_data)

# my expectation
expected_sum_empty_data = pd.DataFrame([],columns=('Project', 'Task', 'Value'))
expected_sum_empty_data.set_index(['Project', 'Task'], inplace=True)
print(expected_sum_empty_data)

Expected Output

Installed Version

python : 3.8.8
pandas : 1.2.3
numpy : 1.20.1
OS: Ubuntu 20.04.1 LTS (docker)

[rhshadrach]

Thanks for the report! This is a duplicate of #26411 which was fixed and will be released in 1.3. Running on master:

empty_data = pd.DataFrame([],columns=('Project', 'Task', 'Increasing Length', 'Start Period', 'Period', 'Value'))
empty_data.set_index(['Project', 'Task', 'Increasing Length', 'Start Period', 'Period'], inplace=True)
sum_empty_data = empty_data.sum(level=['Project', 'Task'])
print(sum_empty_data.index)
print(sum_empty_data.columns)

gives

MultiIndex([], names=['Project', 'Task'])
Index(['Value'], dtype='object')

Reply here if I've missed something and can reopen.