Grouping by all columns of an empty DataFrame should produce MultiIndex, but doesn't

[shwina]

Greetings!

This comes up in cuDF in our tests that check for compatibility with Pandas:

In [21]: a = pd.DataFrame({'a': [], 'b': [], 'c': []})

In [22]: a.groupby(['a', 'b']).sum().index
Out[22]: MultiIndex([], names=['a', 'b'])

In [23]: a.groupby(['a', 'b', 'c']).sum().index
Out[23]: Index([], dtype='object')

Would it be more consistent for grouping by multiple columns to always result in a MultiIndex?

Edit: Checked with Pandas 1.0.1 as well as 0.25.3 -- same behaviour.

[mroeschke]

Looks to be okay in master now. Could use a test

In [3]: In [21]: a = pd.DataFrame({'a': [], 'b': [], 'c': []})

In [4]: In [22]: a.groupby(['a', 'b']).sum().index
Out[4]: MultiIndex([], names=['a', 'b'])

In [5]: In [23]: a.groupby(['a', 'b', 'c']).sum().index
Out[5]: MultiIndex([], names=['a', 'b', 'c'])

[gabrieldi95]

take

[YagoGG]

@gabrieldi95 with your permission, I will be retaking this. Cheers!

[YagoGG]

take