I have multiple Pandas datetime-indexed dataframes, where the index frequency for some dataframes is constant throughout, but for others the frequency switches at some unknown point. For example, the frequency may switch from 5-minute to 1-minute, or from 15-minute to 1-hour.
How can I identify the different frequencies present in the index, and return the start and end times for each?
Here is an example dataframe. Note the switch from 5-minute frequency to 1-minute frequency at 1997-01-01 09:30:00 (magnitude of var_1 happens to change here also, but this is arbitrary).
var_1
Datetime
1997-01-01 07:00:00 10.28
1997-01-01 07:05:00 11.00
1997-01-01 07:10:00 11.06
1997-01-01 07:15:00 11.06
1997-01-01 07:20:00 10.89
1997-01-01 07:25:00 11.39
1997-01-01 07:30:00 11.78
1997-01-01 07:35:00 11.83
1997-01-01 07:40:00 11.94
1997-01-01 07:45:00 12.06
1997-01-01 07:50:00 12.39
1997-01-01 07:55:00 12.61
1997-01-01 08:00:00 12.33
1997-01-01 08:05:00 12.28
1997-01-01 08:10:00 12.44
1997-01-01 08:15:00 12.50
1997-01-01 08:20:00 12.44
1997-01-01 08:25:00 12.11
1997-01-01 08:30:00 11.78
1997-01-01 08:35:00 11.61
1997-01-01 08:40:00 11.50
1997-01-01 08:45:00 11.56
1997-01-01 08:50:00 11.72
1997-01-01 08:55:00 11.67
1997-01-01 09:00:00 11.56
1997-01-01 09:05:00 11.56
1997-01-01 09:10:00 11.56
1997-01-01 09:15:00 11.11
1997-01-01 09:20:00 11.11
1997-01-01 09:25:00 11.33
1997-01-01 09:30:00 2.63
1997-01-01 09:31:00 2.62
1997-01-01 09:32:00 2.58
1997-01-01 09:33:00 2.62
1997-01-01 09:34:00 2.67
1997-01-01 09:35:00 2.68
1997-01-01 09:36:00 2.65
1997-01-01 09:37:00 2.57
1997-01-01 09:38:00 2.43
1997-01-01 09:39:00 2.31
1997-01-01 09:40:00 2.21
1997-01-01 09:41:00 2.13
1997-01-01 09:42:00 2.04
1997-01-01 09:43:00 1.94
1997-01-01 09:44:00 1.82
1997-01-01 09:45:00 1.75
1997-01-01 09:46:00 1.72
1997-01-01 09:47:00 1.71
1997-01-01 09:48:00 1.64
1997-01-01 09:49:00 1.59
1997-01-01 09:50:00 1.54
1997-01-01 09:51:00 1.47
1997-01-01 09:52:00 1.39
1997-01-01 09:53:00 1.31
1997-01-01 09:54:00 1.24
1997-01-01 09:55:00 1.20
1997-01-01 09:56:00 1.19
1997-01-01 09:57:00 1.18
1997-01-01 09:58:00 1.16
1997-01-01 09:59:00 1.13
Ideally, I would like to return something like:
freq1 = ['1997-01-01 07:00:00', '1997-01-01 09:30:00', '5T']
freq2 = ['1997-01-01 09:30:00', '1997-01-01 09:59:00', '1T']
Where the list items are [start time, end time, frequency].
My goal is take any dataframe with unknown frequency distribution, and return the time periods for each frequency that is present.
Here's one option. Create a Series that increments when the frequency changes, and use that to form groups.
df = df.reset_index()
s = df.Datetime.diff().bfill().ne(df.Datetime.diff().shift(-1).ffill()).cumsum()
# Only so the name isn't <lambda>
def freq(x):
return pd.infer_freq(x)
freqs = df.groupby(s).Datetime.agg([min, max, freq]).rename_axis(None, 0)
freqs['max'].update(freqs['min'].shift(-1)) # Fix the end times
min max freq
0 1997-01-01 07:00:00 1997-01-01 09:30:00 5T
1 1997-01-01 09:30:00 1997-01-01 09:59:00 T