Grouping over time in Python

group_by_dynamic buckets rows by a time window, like a resample.

import polars as pl

frame = pl.DataFrame({
    "at": pl.datetime_range(
        pl.datetime(2026, 7, 27), pl.datetime(2026, 8, 2),
        interval="1d", eager=True),
    "tpm": [88, 91, 97, 104, 99, 108, 102],
}).sort("at")

weekly = frame.group_by_dynamic("at", every="3d").agg(
    pl.len().alias("runs"),
    pl.col("tpm").mean().round(1).alias("mean_tpm"),
)
print(weekly)

rolling = frame.with_columns(
    pl.col("tpm").rolling_mean(window_size=3).round(1).alias("rolling3"))
print(rolling.tail(3))

How it works

The frame must be sorted on the time column.
every is the bucket size; period can overlap them.
Empty windows appear or not depending on the arguments.

Keywords and builtins used here

as
print

The run, in numbers

Lines: 18
Characters to type: 470
Tokens: 186
Three-star pace: 110 tpm

At the three-star pace of 110 tokens a minute, this run takes about 101 seconds.

Type this snippet

Step 2 of 2 in Time series, step 24 of 32 in Polars.

← Previous Next →