How to groupby pandas time-series in interval from now backwards?

Question

I have a financial time-series in 15 minute intervals.

2022-01-20 03:45:00+00:00    41941.738281
2022-01-20 04:00:00+00:00    41952.324219
2022-01-20 04:15:00+00:00    41945.421875
2022-01-20 04:30:00+00:00    41921.039062
2022-01-20 04:45:00+00:00    41910.382812
2022-01-20 05:00:00+00:00    41921.597656
2022-01-20 05:15:00+00:00    41914.296875
2022-01-20 05:30:00+00:00    41839.437500
2022-01-20 05:45:00+00:00    41873.421875
2022-01-20 06:00:00+00:00    41905.511719
2022-01-20 06:15:00+00:00    41958.675781
2022-01-20 06:30:00+00:00    42010.332031
2022-01-20 06:45:00+00:00    42045.492188
2022-01-20 07:00:00+00:00    42034.050781
2022-01-20 07:15:00+00:00    42010.828125
2022-01-20 07:30:00+00:00    41976.238281
2022-01-20 07:45:00+00:00    42078.578125
2022-01-20 08:00:00+00:00    42071.261719
2022-01-20 08:15:00+00:00    41995.722656
2022-01-20 08:26:00+00:00    41953.441406
Name: Open, dtype: float64

I want a summary of 6 hour intervals. I want the intervals to begin now. So first interval is from now until now - 6 hours.

I know about Grouper and the freq attribute:

df.groupby(pd.Grouper(level=0, freq="6h")).sum()

However this brings fixed intervals:

2022-01-19 12:00:00+00:00    1.011087e+06
2022-01-19 18:00:00+00:00    1.005388e+06
2022-01-20 00:00:00+00:00    1.005184e+06
2022-01-20 06:00:00+00:00    4.620401e+05

The date_range function can produce intervals I want.

pd.date_range(end="now", periods=4, freq="6h")

out:

DatetimeIndex(['2022-01-19 15:41:08.881073', '2022-01-19 21:41:08.881073',
           '2022-01-20 03:41:08.881073', '2022-01-20 09:41:08.881073'],
          dtype='datetime64[ns]', freq='6H')

How could I group by them.

Answer 1

Use parameter origin in resample for starting time from r range like first value, if necessary filter by maximal (last) value of df.index before:

r = pd.date_range(end="now", periods=4, freq="6h")
df[df.index <= r[-1]].resample('6h', origin=r[0]).sum()