MultiIndex reindex should behave like Index.

[rockg]

xref #10347

I'm having trouble understand how exactly reindex is supposed to work in the case of MultiIndex. What I would like is the exact behavior in the Index case where values can be filled or padded according to some level of the MultiIndex. I can't understand what reindex is doing in the MultiIndex case. The below is an example where I would like to interpolate using the prior value a month frame to an hourly frame.

import pandas as pd
d1 = pd.date_range('1/1/2012', '3/1/2012', freq='MS')
df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]}, index=d1)
dfu = df.unstack()

dfu
Out[12]: 
A  2012-01-01    1
   2012-02-01    2
   2012-03-01    3
B  2012-01-01    4
   2012-02-01    5
   2012-03-01    6
dtype: int64

d2 = pd.date_range('1/1/2012', '3/31/2012', freq='D')

dfu.reindex(d2, level=1) # What is this supposed to do?
Out[17]: 
A  2012-01-01    1
   2012-02-01    2
   2012-03-01    3
B  2012-01-01    4
   2012-02-01    5
   2012-03-01    6
dtype: int64

df.reindex(d2, method='pad').unstack() #Ideally something like this, but without the trickery
Out[19]: 
A  2012-01-01    1
   2012-01-02    1
   2012-01-03    1
   2012-01-04    1
   2012-01-05    1
   2012-01-06    1
   2012-01-07    1
   2012-01-08    1
   2012-01-09    1
   2012-01-10    1
   2012-01-11    1
   2012-01-12    1
   2012-01-13    1
   2012-01-14    1
   2012-01-15    1
...
B  2012-03-17    6
   2012-03-18    6
   2012-03-19    6
   2012-03-20    6
   2012-03-21    6
   2012-03-22    6
   2012-03-23    6
   2012-03-24    6
   2012-03-25    6
   2012-03-26    6
   2012-03-27    6
   2012-03-28    6
   2012-03-29    6
   2012-03-30    6
   2012-03-31    6
Length: 182, dtype: int64

[jreback]

we were just discussing this today
see #7886

[rockg]

That seemed kind of different to me, but maybe I'm not looking closely enough.

[jreback]

see also comments in #7867

[jreback]

you can do this easily with a groupby

something like

df.groupby(level=0).apply(lambda x: x.reindex(d2,method='pad')

prob could directly make reindex a groupby method (so make this syntactically nicer)
maybe have reindex to under the hood

[jreback]

Ya, I guess this is a bit awkward

In [29]: dfu.reset_index().groupby('level_0').apply(lambda x: x.set_index('level_1').reindex(d2,method='pad'))
Out[29]: 
                   level_0  0
level_0                      
A       2012-01-01       A  1
        2012-01-02       A  1
        2012-01-03       A  1
        2012-01-04       A  1
        2012-01-05       A  1
        2012-01-06       A  1
        2012-01-07       A  1
        2012-01-08       A  1
        2012-01-09       A  1
        2012-01-10       A  1
        2012-01-11       A  1
        2012-01-12       A  1
        2012-01-13       A  1
        2012-01-14       A  1
        2012-01-15       A  1
        2012-01-16       A  1
        2012-01-17       A  1
        2012-01-18       A  1
        2012-01-19       A  1
        2012-01-20       A  1
        2012-01-21       A  1
        2012-01-22       A  1
        2012-01-23       A  1
        2012-01-24       A  1
        2012-01-25       A  1
        2012-01-26       A  1
        2012-01-27       A  1
        2012-01-28       A  1
        2012-01-29       A  1
        2012-01-30       A  1
...                    ... ..
B       2012-03-02       B  6
        2012-03-03       B  6
        2012-03-04       B  6
        2012-03-05       B  6
        2012-03-06       B  6
        2012-03-07       B  6
        2012-03-08       B  6
        2012-03-09       B  6
        2012-03-10       B  6
        2012-03-11       B  6
        2012-03-12       B  6
        2012-03-13       B  6
        2012-03-14       B  6
        2012-03-15       B  6
        2012-03-16       B  6
        2012-03-17       B  6
        2012-03-18       B  6
        2012-03-19       B  6
        2012-03-20       B  6
        2012-03-21       B  6
        2012-03-22       B  6
        2012-03-23       B  6
        2012-03-24       B  6
        2012-03-25       B  6
        2012-03-26       B  6
        2012-03-27       B  6
        2012-03-28       B  6
        2012-03-29       B  6
        2012-03-30       B  6
        2012-03-31       B  6

[182 rows x 2 columns]