weights not yet supported on array with null values

Error message

weights not yet supported on array with null values

What it means

nulls::rolling_quantile (the null-handling rolling quantile/percentile kernel) guards against combining weights with nulls: if weights is Some(_) it panics 'weights not yet supported on array with null values'. Weighted quantiles on the null path are unimplemented; the comment in the source also notes the dancing-links implementation is disabled pending a fix.

Source

Thrown at crates/polars-compute/src/rolling/nulls/quantile.rs:141

    params: Option<RollingFnParams>,
) -> ArrayRef
where
    T: NativeType
        + IsFloat
        + Float
        + std::iter::Sum
        + AddAssign
        + SubAssign
        + Div<Output = T>
        + NumCast
        + One
        + Zero
        + SealedRolling
        + PartialOrd
        + Sub<Output = T>,
{
    if weights.is_some() {
        panic!("weights not yet supported on array with null values")
    }
    let offset_fn = match center {
        true => det_offsets_center,
        false => det_offsets,
    };
    /*
    TODO: fix or remove the dancing links based rolling implementation
    see https://github.com/pola-rs/polars/issues/23480
    if !center {
        let params = params.as_ref().unwrap();
        let RollingFnParams::Quantile(params) = params else {
            unreachable!("expected Quantile params");
        };

        let out = super::quantile_filter::rolling_quantile::<_, MutablePrimitiveArray<_>>(
            params.method,
            min_periods,
            window_size,

View on GitHub (pinned to 9b5d73fd00)

Solutions

Drop the weights argument for nullable input - unweighted rolling quantile is supported
Fill nulls first and accept the statistical impact: s.fill_null(...).rolling_quantile(..., weights=...)
Compute on drop_nulls() and reindex where window alignment can be restored
Request weighted null-aware quantiles upstream

Example fix

# before
s.rolling_quantile(0.95, window_size=30, weights=w)  # nulls -> panic
# after
s.fill_null(method="forward").rolling_quantile(0.95, window_size=30, weights=w)

Defensive patterns

Strategy: validation

Validate before calling

def safe_rolling_quantile(s: pl.Series, quantile: float, window_size: int,
                           weights=None, min_periods=1):
    if weights is not None and s.null_count() > 0:
        raise ValueError("weights + nulls unsupported for rolling_quantile")
    return s.rolling_quantile(quantile, window_size, weights=weights, min_periods=min_periods)

Try / catch

try:
    out = s.rolling_quantile(0.95, window_size=30, weights=w)
except pl.exceptions.PanicException:
    out = s.fill_null(method="forward").rolling_quantile(0.95, window_size=30, weights=w)

Prevention

Guard quantile alerting code against null+weights combinations
Forward-fill monitoring series before weighted percentiles
Log input null counts alongside rolling metrics for diagnosability

When it happens

Trigger: s.rolling_quantile(quantile=0.9, window_size=k, weights=[...]) on a Series with at least one null; rolling median/percentile with exponential decay weights over nullable metrics.

Common situations: Monitoring/SLO pipelines computing weighted rolling p95 over series with missing samples; quantile alerting that works on healthy data and panics during partial outages when nulls appear.

Related errors

AI-assisted analysis of pola-rs/polars@9b5d73fd00 (2026-08-19). Data as JSON: /api/errors/7ea0a8a07a29f039. Report an issue: GitHub.