ENH: allow datelike to be aggregated in resample

[jreback]

http://stackoverflow.com/questions/21878599/pandas-drops-timestamp-columns-in-resample

init_time=pd.date_range(start='2010-01-01 00:00', end='2010-01-05 00:00', freq='12H')
valid_time=pd.date_range(start='2010-01-01 00:00', end='2010-01-05 00:00', freq='12H')
data = np.random.random(len(valid_time))
frame = pd.DataFrame(index=valid_time, data=data)
frame['init_time'] = init_time

Works, but a little hacky

In [48]: numeric = frame.resample('D',how='mean')

In [49]: datelike = frame.loc[:,frame.dtypes.isin([np.dtype('datetime64[ns]')])]

In [50]: datelike = datelike.resample('D',
         how=lambda x: Timestamp(x.view('i8').mean()))

In [51]: concat([numeric,datelike],axis=1)
Out[51]: 
                   0           init_time
2010-01-01  0.798880 2010-01-01 06:00:00
2010-01-02  0.859781 2010-01-02 06:00:00
2010-01-03  0.515503 2010-01-03 06:00:00
2010-01-04  0.505557 2010-01-04 06:00:00
2010-01-05  0.979835 2010-01-05 00:00:00

[5 rows x 2 columns]

[jbrockmendel]

frame.resample("D").mean() now retains the dt64 column. i think this is closable?

[MarcoGorelli]

yeah this seems to work now

In [24]: frame
Out[24]:
                            0           init_time
2010-01-01 00:00:00  0.110461 2010-01-01 00:00:00
2010-01-01 12:00:00  0.063558 2010-01-01 12:00:00
2010-01-02 00:00:00  0.016708 2010-01-02 00:00:00
2010-01-02 12:00:00  0.685998 2010-01-02 12:00:00
2010-01-03 00:00:00  0.607047 2010-01-03 00:00:00
2010-01-03 12:00:00  0.090777 2010-01-03 12:00:00
2010-01-04 00:00:00  0.871941 2010-01-04 00:00:00
2010-01-04 12:00:00  0.928479 2010-01-04 12:00:00
2010-01-05 00:00:00  0.272012 2010-01-05 00:00:00

In [25]: frame.resample('D').mean()
Out[25]:
                   0           init_time
2010-01-01  0.087010 2010-01-01 06:00:00
2010-01-02  0.351353 2010-01-02 06:00:00
2010-01-03  0.348912 2010-01-03 06:00:00
2010-01-04  0.900210 2010-01-04 06:00:00
2010-01-05  0.272012 2010-01-05 00:00:00

closing then