Function compute_pca

crates/openquant/src/feature_importance.rs:207–246

(
    feature_rows: &[Vec<f64>],
    variance_thresh: f64,
)

    let inv_rank: Vec<f64> = pca_rank.iter().map(|r| 1.0 / *r as f64).collect();
    let weighted = weighted_kendall_tau(feature_importance_mean, &inv_rank);

    Ok(PcaCorrelation { pearson, spearman, kendall, weighted_kendall_rank: weighted })
}

pub fn plot_feature_importance(
    importance: &BTreeMap<String, ImportanceStats>,
    oob_score: f64,
    oos_score: f64,
    output_path: Option<&str>,
) -> Result<(), FeatureImportanceError> {
    if let Some(path) = output_path {
        let mut s = format!("oob_score,{oob_score}\noos_score,{oos_score}\nfeature,mean,std\n");
        for (k, v) in importance {
            s.push_str(&format!("{k},{},{}\n", v.mean, v.std));
        }
        std::fs::write(path, s).map_err(|e| FeatureImportanceError::WriteOutput(e.to_string()))?;
    }
    Ok(())
}

/// PCA output: `(eigenvalues, eigenvectors, standardized feature rows)`.
type PcaDecomposition = (Vec<f64>, DMatrix<f64>, Vec<Vec<f64>>);

fn compute_pca(
    feature_rows: &[Vec<f64>],
    variance_thresh: f64,
) -> Result<PcaDecomposition, FeatureImportanceError> {
    if feature_rows.iter().any(|r| r.len() != feature_rows[0].len()) {
        return Err(FeatureImportanceError::RaggedFeatureRows);
    }
    let x_std = standardize(feature_rows);
    let x = to_dmatrix(&x_std);
    let dot = x.transpose() * &x;
    let eig = SymmetricEigen::new(dot);

    let mut idx: Vec<usize> = (0..eig.eigenvalues.len()).collect();
    idx.sort_by(|&a, &b| {
        eig.eigenvalues[b].partial_cmp(&eig.eigenvalues[a]).unwrap_or(std::cmp::Ordering::Equal)
    });
    let mut eval = Vec::with_capacity(idx.len());
    let mut evec_cols = Vec::with_capacity(idx.len());
    for i in idx {
        eval.push(eig.eigenvalues[i]);