pola-rs/polars · error

activate 'propagate_nans'

Error message

activate 'propagate_nans'

What it means

This panic is raised by spearman_rank_corr when the polars Python package was compiled without the 'propagate_nans' feature enabled. The Rust binding only contains the real implementation behind #[cfg(feature = "propagate_nans")]; without it, the function unconditionally panics with a directive telling the user to activate the feature.

Solutions

1. Install the official polars wheel from PyPI, which includes this feature.
2. If building from source, rebuild with the propagate_nans feature enabled (cargo build with --features propagate_nans or the 'all-features' option).
3. As a workaround, compute Spearman correlation via rank + pearson corr: pl.corr(a.rank(), b.rank(), method='pearson').

Example fix

# before (custom build without feature)
df.select(pl.corr('a', 'b', method='spearman', propagate_nans=True))
# after: rank-then-pearson workaround, or reinstall official wheel
pip install --upgrade --force-reinstall polars
df.select(pl.corr('a', 'b', method='pearson').fill_nan(None))  # after manual rank

import polars as pl
assert 'propagate_nans' in getattr(pl, '_features', getattr(pl.build_info(), 'features', [])) if hasattr(pl, 'build_info') else True
# or feature-check at runtime:
import polars; assert 'propagate_nans' in str(polars.build_info()), 'propagate_nans feature required'

def supports_propagate_nans():
    import polars
    try:
        return 'propagate_nans' in polars.build_info()['features']
    except Exception:
        return True

try:
    corr = pl.corr('a', 'b', method='spearman', propagate_nans=True)
except BaseException as e:
    if 'propagate_nans' in str(e):
        corr = pl.corr(pl.col('a').rank(), pl.col('b').rank(), method='pearson')
    else:
        raise

When it happens

Trigger: Calling pl.spearman_rank_corr (or the corresponding expression function) on a polars build installed without the propagate_nans cargo feature.

Common situations: Using a minimal/custom-compiled polars build (e.g. from source with a reduced feature set) or an environment where the binary was built without this feature; more common when building polars from source or using alternative distribution channels.

Related errors

• activate 'binary_encoding' feature
• activate dtype-categorical to convert dictionary arrays
• activate 'object' feature
• activate 'timezones' feature
• data type must be FixedSizeList

AI-assisted analysis of pola-rs/polars@fe841f959e (2026-09-18). Data as JSON: /api/errors/a89409fcc039390c. Report an issue: GitHub.

Appendix: source

Thrown at crates/polars-python/src/functions/lazy.rs:586

    let mut value = value.inner;
    let n = n.inner;

    if let Some(dtype) = dtype {
        value = value.cast(dtype.0);
    }

    dsl::repeat(value, n).into()
}

#[pyfunction]
pub fn spearman_rank_corr(a: PyExpr, b: PyExpr, propagate_nans: bool) -> PyExpr {
    #[cfg(feature = "propagate_nans")]
    {
        dsl::spearman_rank_corr(a.inner, b.inner, propagate_nans).into()
    }
    #[cfg(not(feature = "propagate_nans"))]
    {
        panic!("activate 'propagate_nans'")
    }
}

#[pyfunction]
#[cfg(feature = "sql")]
pub fn sql_expr(sql: &str) -> PyResult<PyExpr> {
    let expr = polars::sql::sql_expr(sql).map_err(PyPolarsErr::from)?;
    Ok(expr.into())
}

View on GitHub (pinned to fe841f959e)