[Python] Calculate cumulative sum of time series X for time points in series Y

Imagine transactions, identified by amount, arriving throughout the day. You want to calculate the running total of amount at given points in time (9 am, 10 am, etc.).

With pandas, I would use apply to perform such an operation. With Polars, I tried using map_elements. I have also considered group_by_dynamic but I am not sure it gives me control of the time grid's start / end / increment.

Is there a better way?

import polars as pl
import datetime

df = pl.DataFrame({
"time": [
datetime.datetime(2025, 2, 2, 11, 1),
datetime.datetime(2025, 2, 2, 11, 2),
datetime.datetime(2025, 2, 2, 11, 3)
],
"amount": [5.0, -1, 10]
})

dg = pl.DataFrame(
pl.datetime_range(
datetime.datetime(2025, 2, 2, 11, 0),
datetime.datetime(2025, 2, 2, 11, 5),
"1m",
eager = True
),
schema=["time"]
)

def _cumsum(dt):
return df.filter(pl.col("time") <= dt).select(pl.col("amount")).sum().item()

dg.with_columns(
cum_amount=pl.col("time").map_elements(_cumsum, return_dtype= pl.Float64)
)

Continue reading...