Difference in groupby behavior between Pandas 0.13.1 and 0.15.2

[dakoner]

Hi, I am seeing a difference in behavior on this groupby between Pandas 0.13.1 and 0.15.2. Specifically, it's like 0.15.2 is doing a cross join while 0.13.1 isn't.

print pandas.DataFrame([
  {'a': 1, 'b': 2, 'c': 3},
  {'a': 4, 'b': 5, 'c': 6}, ]).set_index(
    list('ab')).groupby(level=list('ab')).mean()

0.13.1 produces:

     c
a b   
1 2  3
4 5  6
[2 rows x 1 columns]

while 0.15.2 produces

      c
a b    
1 2   3
  5 NaN
4 2 NaN
  5   6

basically, the same matrix, but with extra cross NaN entries.

We're wondering if this behavior is intentional, or a bug. It wasn't entirely clear from the set of release notes that the groupby behavior changed so much.

[jreback]

the resulting resulting cartesian product of the indices (e.g. the nan entries), were a bug in 0.15+ (maybe only in 0.15.2), and are fixed in 0.16.0 (coming soon), fixed in #9177