Pandas Feature Request: groupby retaining indexes

The feature I'd like to suggest/request is pretty simple, it has workarounds but it seems very odd to me that this is not a native behavior. Specially because the workarounds seem so slow.

I imagine it as a simple parameter for the groupby function.

The feature is: the ability to group a level (of a MultiIndex dataframe) but still retain that level's labels. I think it will be clear with the example below.

Note that this came from a question I asked in stack overflow

Code Sample

df = pd.DataFrame([['A', 1, 3],
                   ['A', 2, 4],
                   ['A', 3, 6],
                   ['B', 1, 9],
                   ['B', 2, 10],
                   ['B', 4, 6]],
                  columns=pd.Index(['Name', 'Date', 'Value'], name='ColumnName')
                 ).set_index(['Name', 'Date'])

df.groupby(level=('Name')).last()

Intended Output

ColumnName         Value
Name    Date
A        3           6
B        4           6

output of pandas 0.18.0

ColumnName         Value
Name    
A                    6
B                    6

Comment From: jorisvandenbossche

I also just answered on SO, but the solution is rather simple: when doing a groupby().last(), the last item for each column will be calculated. So, you can just reset the index level to make a column of it, and then do the groupby:

In [44]: df.reset_index(level='Date').groupby(level=0).last()
Out[44]:
ColumnName  Date  Value
Name
A              3      6
B              4      6

Comment From: pedrobraz1990

Thanks @jorisvandenbossche , yet you don't think that a native way (faster) is justifiable ?

Comment From: jreback

@pedrobraz1990 this is as fast as it gets.

These are all cheap operations. If you have a performance issue then you are doing things in a non-performant / non-idiomatic way, e.g. iterating.

Comment From: pedrobraz1990

all right then. should I close this issue request then ?

Comment From: jreback

already did

Comment From: jorisvandenbossche

BTW, there is actually an even easier solution:

In [202]: df.groupby(level='Name').tail(1)
Out[202]:
ColumnName  Value
Name Date
A    3          6
B    4          6

.tail() acts like a 'filter' method, so keeping the index intact.

Comment From: pedrobraz1990

but then I won't get the whole dataframe, right ?

@jorisvandenbossche

Comment From: jorisvandenbossche

@pedrobraz1990 What do you mean exactly? Of course you don't get the whole dataframe, as you are selecting the last row of each group?

Comment From: pedrobraz1990

oh sorry. I had the impression that tail only returned the last 5 rows. But yes, I meant all the last of each group rows.

Comment From: jorisvandenbossche

Ah yes, tail() on a dataframe will indeed return the last 5 rows, but when called on a groupby object, it will do that for each group (so tail(1) only the last 1 row for each group).

Comment From: pedrobraz1990

Awesome ! thanks a lot @jorisvandenbossche