[Python] Polars - rolling - How to reset count each day?

I've switched from Pandas-Python to Polars and I'm trying to figure out how to essentially do a rolling sum for each day. I don't want just the total for each day, I want the totals for each time period (5m period) each day, in a "rolling_sum" fashion. Here is the code I have so far:

from datetime import datetime, timedelta

df = pl.DataFrame({
"date": pl.datetime_range(
datetime(1985, 1, 1),
datetime(1985, 1, 6),
timedelta(minutes=5),
time_unit="ns",
eager=True
)
})

df = df.with_columns(
A=pl.lit(2, dtype=pl.Int32)
)

df = (
df.lazy()
.rolling("date", period="1d")
.agg(
pl.col("A").sum().alias("A_daily")
)
.collect()
)

print(df[280:])


This produces:

shape: (1_161, 2)
┌─────────────────────┬─────────┐
│ date ┆ A_daily │
│ --- ┆ --- │
│ datetime[ns] ┆ i32 │
╞═════════════════════╪═════════╡
│ 1985-01-01 23:20:00 ┆ 562 │
│ 1985-01-01 23:25:00 ┆ 564 │
│ 1985-01-01 23:30:00 ┆ 566 │
│ 1985-01-01 23:35:00 ┆ 568 │
│ 1985-01-01 23:40:00 ┆ 570 │
│ 1985-01-01 23:45:00 ┆ 572 │
│ 1985-01-01 23:50:00 ┆ 574 │
│ 1985-01-01 23:55:00 ┆ 576 │
│ 1985-01-02 00:00:00 ┆ 576 │
│ 1985-01-02 00:05:00 ┆ 576 │
│ 1985-01-02 00:10:00 ┆ 576 │
│ 1985-01-02 00:15:00 ┆ 576 │
│ 1985-01-02 00:20:00 ┆ 576 │
│ 1985-01-02 00:25:00 ┆ 576 │
│ 1985-01-02 00:30:00 ┆ 576 │
│ 1985-01-02 00:35:00 ┆ 576 │
│ 1985-01-02 00:40:00 ┆ 576 │
│ 1985-01-02 00:45:00 ┆ 576 │
│ 1985-01-02 00:50:00 ┆ 576 │
...
│ 1985-01-05 23:50:00 ┆ 576 │
│ 1985-01-05 23:55:00 ┆ 576 │
│ 1985-01-06 00:00:00 ┆ 576 │
└─────────────────────┴─────────┘


What I guess what I'm actually after is for each day, I want the A_Daily to reset, i.e. to look like:

shape: (1_161, 2)
┌─────────────────────┬─────────┐
│ date ┆ A_daily │
│ --- ┆ --- │
│ datetime[ns] ┆ i32 │
╞═════════════════════╪═════════╡
│ 1985-01-01 23:20:00 ┆ 562 │
│ 1985-01-01 23:25:00 ┆ 564 │
│ 1985-01-01 23:30:00 ┆ 566 │
│ 1985-01-01 23:35:00 ┆ 568 │
│ 1985-01-01 23:40:00 ┆ 570 │
│ 1985-01-01 23:45:00 ┆ 572 │
│ 1985-01-01 23:50:00 ┆ 574 │
│ 1985-01-01 23:55:00 ┆ 576 │
│ 1985-01-02 00:00:00 ┆ 2 │ # !!!! CHANGE HERE !!!!
│ 1985-01-02 00:05:00 ┆ 4 │
│ 1985-01-02 00:10:00 ┆ 6 │
│ 1985-01-02 00:15:00 ┆ 8 │
│ 1985-01-02 00:20:00 ┆ 10 │
│ 1985-01-02 00:25:00 ┆ 12 │
│ 1985-01-02 00:30:00 ┆ 14 │
│ 1985-01-02 00:35:00 ┆ 18 │
│ 1985-01-02 00:40:00 ┆ 20 │
│ 1985-01-02 00:45:00 ┆ 22 │
│ 1985-01-02 00:50:00 ┆ 24 │



Basically, I want to just reset the rolling sum each day. In Pandas, I'd have done it sort of like this:

tn = outdata.groupby(pd.Grouper(freq='D', key='_daily_reset'))[
# '_vp_volume'].transform('cumsum')


Thanks!

Continue reading...