Group by on time series in polars

group_by_dynamic

Group by operations are powerful in Polars, but when working with time series data, you can make them even more effective by using group_by_dynamic instead of the standard group_by. Below is a dataframe showing power usage recorded every half hour over five days.

shape: (193, 2)

Time	Power_Usage
datetime[ms]	i64
2025-02-01 00:00:00	2
2025-02-01 00:30:00	1
2025-02-01 01:00:00	5
2025-02-01 01:30:00	4
2025-02-01 02:00:00	4
…	…
2025-02-04 22:00:00	2
2025-02-04 22:30:00	4
2025-02-04 23:00:00	4
2025-02-04 23:30:00	2
2025-02-05 00:00:00	6

Aggregate on time series

Unlike the standard group_by, which limits you to predefined datetime components such as month or year, group_by_dynamic allows you to aggregate data over custom durations. In this example, the data is aggregated every five and a half hours.

(df
1 .sort('Time')
 .group_by_dynamic('Time', every='5h30m')
 .agg(pl.sum('Power_Usage'))
 )

1

The data must be sorted before using group_by_dynamic, otherwise the code throws an error.

shape: (18, 2)

Time	Power_Usage
datetime[ms]	i64
2025-01-31 22:00:00	21
2025-02-01 03:30:00	59
2025-02-01 09:00:00	61
2025-02-01 14:30:00	50
2025-02-01 20:00:00	64
…	…
2025-02-03 21:30:00	65
2025-02-04 03:00:00	60
2025-02-04 08:30:00	48
2025-02-04 14:00:00	73
2025-02-04 19:30:00	56

Enroll to learn more in my Polars course.