Pandas time series: groupby and sum from noon to noon - python-2.7

Question

My pandas dataframe is structured like this (with 'date' as index):
starttime duration_seconds
date
2012-12-24 11:52:00 31800
2012-12-23 0:28:00 35940
2012-12-22 2:00:00 26820
2012-12-21 1:57:00 23520
2012-12-20 1:32:00 23100
2012-12-19 0:50:00 25080
2012-12-18 1:17:00 24780
2012-12-17 0:38:00 25440
2012-12-15 10:38:00 32760
2012-12-14 0:35:00 23160
2012-12-12 22:54:00 3960
2012-12-12 0:21:00 24060
2012-12-10 23:45:00 900
2012-12-11 11:00:00 24840
2012-12-10 0:27:00 25980
2012-12-09 19:29:00 4320
2012-12-09 3:00:00 29880
2012-12-08 2:07:00 34380
I use the following to groupby date and sum the total seconds each day:
df_sum = df.groupby(df.index.date).sum()
What I'd like to do is sum duration_seconds from noon on one day to noon on the following day. Is there an elegant (pandas) way of doing this? Thanks in advance!

Answer 1

pd.TimeGrouper is a custom groupby class for time-interval grouping of NDFrames with a DatetimeIndex, TimedeltaIndex or PeriodIndex. (If your dataframe index is using date-strings, you'll need to convert it to a DatetimeIndex first by using df.index = pd.DatetimeIndex(df.index).)
df.groupby(pd.TimeGrouper('24H')).sum() groups df using 24-hour intervals starting at time 00:00:00.
df.groupby(pd.TimeGrouper('24H'), base=12).sum() groups df using 24-hour intervals starting at time 12:00:00:
In [90]: df.groupby(pd.TimeGrouper('24H', base=12)).sum()
Out[90]:
duration_seconds
2012-12-07 12:00:00 34380.0
2012-12-08 12:00:00 34200.0
2012-12-09 12:00:00 26880.0
2012-12-10 12:00:00 24840.0
2012-12-11 12:00:00 28020.0
2012-12-12 12:00:00 NaN
2012-12-13 12:00:00 23160.0
2012-12-14 12:00:00 32760.0
2012-12-15 12:00:00 NaN
2012-12-16 12:00:00 25440.0
2012-12-17 12:00:00 24780.0
2012-12-18 12:00:00 25080.0
2012-12-19 12:00:00 23100.0
2012-12-20 12:00:00 23520.0
2012-12-21 12:00:00 26820.0
2012-12-22 12:00:00 35940.0
2012-12-23 12:00:00 31800.0
Documentation on pd.TimeGrouper is a little sparse. It is a subclas of pd.Grouper and thus many of its parameters have the same meaning as those documented for pd.Grouper. You can find more examples of pd.TimeGrouper usage in the Cookbook. I found the base parameter by inspecting the source code. The base parameter in pd.TimeGrouper has the same meaning as the base parameter in pd.resample and that is not surprising since pd.resample is implemented using pd.TimeGrouper.
In fact, come to think of it, another way to compute the desired result is
df.resample('24H', base=12).sum()