DataFrameGroupBy.cumcount() returing Series instead of DataFrame

[jorisvandenbossche]

@hayd

DataFrameGroupBy.cumcount() returns a Series while most others (the ones I tested, like count, cumsum, first, sum ) return a DataFrame. They are maybe not all comparable, but at least with cumsum it could be I think:

In [1]: data = pd.DataFrame({'name' : ['a', 'a', 'b', 'd'], 'counts' : [3,4,3,2]})
In [2]: data
Out[2]:
   counts name
0       3    a
1       4    a
2       3    b
3       2    d

In [3]: g = data.groupby('name')
In [4]: g.cumcount()
Out[4]:
0    0
1    1
2    0
3    0
dtype: int64

In [5]: g.count()
Out[5]:
      counts  name
name
a          2     2
b          1     1
d          1     1

In [6]: g.first()
Out[6]:
      counts
name
a          3
b          3
d          2

In [7]: g.sum()
Out[7]:
      counts
name
a          7
b          3
d          2

In [8]: g.cumsum()
Out[8]:
  counts name
0      3    a
1      7   aa
2      3    b
3      2    d

[hayd]

I disagree, cumcount is not "looking" at the columns.

• It makes sense for cumsum, sum, etc. since they are summing each column.
• Similarly for first, it's the first "row" (without the as_index)
• count with the count in each column (note it excludes NaN).

[jorisvandenbossche]

Ah, yes. Indeed, that way it is logical!
And also g.size() does return a Series, which is similar (and because of the g.count being equal for all columns, I didn't thought of NaNs influencing the result).

[hayd]

Thanks for reporting anyway, made me noticed some low-hanging fruit here #5614 (I think cumsum etc.? is a bit wrong)

[jorisvandenbossche]

Yes indeed, I wanted to bring that up also, but saw that you already did it :-)