feat: Add rolling_(min/max/sum), rolling_(min/max/sum)_by for pl.Boolean

[EvanLai88]

On version 1.9.0, rolling_(min/max/sum), rolling_(min/max/sum)_by operations are not implemented for dtype pl.Boolean.

Before:

>>> import polars as pl
>>> pl.DataFrame(
...     {"a": [-10, 34, 20, -30, -20, -45, 30]},
... ).with_columns(
...     true_cnt_in_2_win=pl.col("a").gt(0).rolling_sum(2),
...     all_true_in_2_win=pl.col("a").gt(0).rolling_min(2),
...     any_true_in_2_win=pl.col("a").gt(0).rolling_max(2),
... )
thread 'polars-0' panicked at /home/runner/work/polars/polars/crates/polars-time/src/chunkedarray/rolling_window/dispatch.rs:209:9:
not implemented for dtype Boolean
thread 'polars-1' panicked at /home/runner/work/polars/polars/crates/polars-time/src/chunkedarray/rolling_window/dispatch.rs:281:9:
not implemented for dtype Boolean
thread 'polars-2' panicked at /home/runner/work/polars/polars/crates/polars-time/src/chunkedarray/rolling_window/dispatch.rs:319:9:
not implemented for dtype Boolean

After:

>>> import polars as pl
>>> pl.DataFrame(
...     {"a": [-10, 34, 20, -30, -20, -45, 30]},
... ).with_columns(
...     true_cnt_in_2_win=pl.col("a").gt(0).rolling_sum(2),
...     all_true_in_2_win=pl.col("a").gt(0).rolling_min(2),
...     any_true_in_2_win=pl.col("a").gt(0).rolling_max(2),
... )
shape: (7, 4)
┌─────┬───────────────────┬───────────────────┬───────────────────┐
│ a   ┆ true_cnt_in_2_win ┆ all_true_in_2_win ┆ any_true_in_2_win │
│ --- ┆ ---               ┆ ---               ┆ ---               │
│ i64 ┆ u32               ┆ bool              ┆ bool              │
╞═════╪═══════════════════╪═══════════════════╪═══════════════════╡
│ -10 ┆ null              ┆ null              ┆ null              │
│ 34  ┆ 1                 ┆ false             ┆ true              │
│ 20  ┆ 2                 ┆ true              ┆ true              │
│ -30 ┆ 1                 ┆ false             ┆ true              │
│ -20 ┆ 0                 ┆ false             ┆ false             │
│ -45 ┆ 0                 ┆ false             ┆ false             │
│ 30  ┆ 1                 ┆ false             ┆ true              │
└─────┴───────────────────┴───────────────────┴───────────────────┘

Note:
Other rolling_* operations, such as rolling_mean, already support the pl.Boolean dtype via to_float(), if I’m not mistaken.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.65%. Comparing base (194b31e) to head (46cf22c).

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #19112   +/-   ##
=======================================
  Coverage   79.65%   79.65%           
=======================================
  Files        1531     1531           
  Lines      208937   208956   +19     
  Branches     2418     2418           
=======================================
+ Hits       166421   166442   +21     
+ Misses      41969    41967    -2     
  Partials      547      547           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ritchie46]

This should cast to IdxType.

[ritchie46]

I really don't think we should cast here, but rather have a specialize algorithm for booleans.

[ritchie46]

I really don't think we should cast here, but rather have a specialize algorithm for booleans.

[ritchie46]

I really don't think we should cast here, but rather have a specialize algorithm for booleans.