BUG/ENH: groupby with a list of customgroup and string should work

[jreback]

see #3791, #2450

The following should work (but raises now)

import datetime as DT

df = pd.DataFrame({
'Branch' : 'A A A A A B'.split(),
'Buyer': 'Carl Mark Carl Joe Joe Carl'.split(),
'Quantity': [1,3,5,8,9,3],
'Date' : [
DT.datetime(2013,1,1,13,0),
DT.datetime(2013,1,1,13,5),
DT.datetime(2013,10,1,20,0),
DT.datetime(2013,10,2,10,0),
DT.datetime(2013,12,2,12,0),                                      
DT.datetime(2013,12,2,14,0),
]})

df = df.set_index('Date', drop=False)
df.groupby([TimeGrouper('6M'),'Buyer']).sum()

[cpcloud]

@jreback forgive me for not reading the threads, but why not just do a resample then groupby buyer?

[jreback]

because you don't want to actually resample (and reduce the time groups), rather you really want a grouping by time, then a sub-grouping by the Buyer in this case, then apply the function

and end up with a multi-level index

[jreback]

you CAN do it with a nested how function (or actually that might be a bug too)

[cpcloud]

ah i c. duh.

[cpcloud]

i think what i was thinking of is df.groupby(TimeGrouper('6M')).sum().groupby('Buyer').sum()

[jreback]

that doesn't work because the string columns are not propogated (I think if you did a custom how instead of sum you could make it work though)

[cpcloud]

i haven't really thought about the formalisms behind groupby but are reductions supposed to distribute across them in general, i.e., df.gb(grp1).reduc.gb(grp2).reduc == df.gb([grp1, grp2]).reduc (pseudocode)

[cpcloud]

if f_1, f_2, ..., f_n are grouping functions should u be able to do reduc(f_1 * f_2 * ... * f_n) where * is composition and reduc is a reduction, i guess is what i'm asking...

[jreback]

groupby could be distributive, but it usually not so; if the first operation were a transform (rather than a reduction) then it could be distributive.....

[cpcloud]

ah ok. thanks.

[hayd]

Also, groupbys from TGs seem to have some useful functionality suppressed/hidden (?) made separate issue

[jreback]

@hayd maybe make a sep issue and link to this one

FYI the so question this came from (using timegrouper and between time)
use new groupby filter method with the between time as your discriminator!