ENH: groupby aggregate with multi-level columns

[jorisvandenbossche]

See http://stackoverflow.com/questions/28833074/aggregate-group-with-multi-level-columns

So we want to have some built-in functionality for this?

The example:

import itertools
import pandas as pd

lev1 = ['foo', 'bar', 'baz']
lev2 = list('abc')

n = 6

df = pd.DataFrame({k: np.random.randn(n) for k in itertools.product(lev1,lev2)}, 
                  index=pd.DatetimeIndex(start='2015-01-01', periods=n, freq='11D'))

             bar               baz               foo            
               a     b     c     a     b     c     a     b     c
2015-01-01 -1.11  2.12 -1.00  0.18  0.14  1.24  0.73  0.06  3.66
2015-01-12 -1.43  0.75  0.38  0.04 -0.33 -0.42  1.00 -1.63 -1.35
2015-01-23  0.01 -1.70 -1.39  0.59 -1.10 -1.17 -1.51 -0.54 -1.11
2015-02-03  0.93  0.70 -0.12  1.07 -0.97 -0.45 -0.19  0.11 -0.79
2015-02-14  0.30  0.49  0.60 -0.28 -0.38  1.11  0.15  0.78 -0.58
2015-02-25 -0.26  0.51  0.82  0.05 -1.45  0.14  0.53 -0.33 -1.35

The question is here if it should be possible in groupby().aggregate() to specify that you want to apply a function to all columns of a certain level label.
E.g. df.groupby(pd.TimeGrouper('MS')).aggregate({'bar': np.sum, 'baz': np.mean, 'foo': np.min}) does not work at the moment.

Or does this lead to far?

[jreback]

xref of #9052, #8593
we need a way to specify how to do this, something like pd.Summary(....)

what you are doing above should work, but I think is just NotImplemented ATM