[Python] Interpretation of nan in rolling_mean with polars

How can we do to do a rolling mean with ignoring the np.nan values that would be in it. From what I read on the documentation, I don t get if it is not possible at all or if some tricky parameters must be passed. If it is not implemented at all, it would be a great idea to add it because it prevents from doing commands such as "do the rolling mean on the N previous row, and ignoring the rows that does not follow a certain condition in those previous N rows".

Here is a quick example to show my problem/question.

import polars as pl
import numpy as np
test = np.array([1, 2, np.nan, 3, 4, np.nan]).reshape(-1, 1)
test = pl.DataFrame(test, schema=["a"])
test.with_columns(rolling_mean=pl.col("a").rolling_mean(window_size=2))


shape: (6, 2)
┌─────┬──────────────┐
│ a ┆ rolling_mean │
│ --- ┆ --- │
│ f64 ┆ f64 │
╞═════╪══════════════╡
│ 1.0 ┆ null │
│ 2.0 ┆ 1.5 │
│ NaN ┆ NaN │
│ 3.0 ┆ NaN │
│ 4.0 ┆ 3.5 │
│ NaN ┆ NaN │
└─────┴──────────────┘

Continue reading...