More flexible multiple function application with DataFrameGroupBy

[wesm]

cc @arthurgerigk

Something of this nature would be nice:

df.groupby(['A', 'B']).agg({'C' : [np.mean, np.std]})

Would also be nice to add some options to agg for what to do with the input dict (by default targets columns)

[wesm]

Here's the outcome:

In [5]: df
Out[5]: 
     A      B         C         D
0  foo    one  0.727964  0.406402
1  bar    one  1.653679  0.266521
2  foo    two -1.677686 -1.143690
3  bar  three -0.591334 -0.594644
4  foo    two -0.839865  0.153575
5  bar    two -0.857455 -0.054324
6  foo    one -0.148167  1.080535
7  foo  three  0.118469  0.715423

In [6]: grouped = df.groupby('A')

In [7]: result = grouped.aggregate({'C' : [np.mean, np.std],
   ...:                                     'D' : [np.mean, np.std]})

In [8]: result
Out[8]: 
            C                   D          
         mean       std      mean       std
A                                          
bar  0.068297  1.379414 -0.127482  0.435219
foo -0.363857  0.925198  0.242449  0.848860

In [9]: grouped.aggregate({'C' : np.mean, 'D' : [np.mean, np.std]})
Out[9]: 
            C         D          
         mean      mean       std
A                                
bar  0.068297 -0.127482  0.435219
foo -0.363857  0.242449  0.848860

In [10]: paste
        result = grouped.aggregate({'C' : np.mean,
                                    'D' : {'foo': np.mean,
                                           'bar': np.std}})
## -- End pasted text --

In [11]: result
Out[11]: 
            C         D          
         mean       foo       bar
A                                
bar  0.068297 -0.127482  0.435219
foo -0.363857  0.242449  0.848860