sklears-cross-decomposition

Latest release: 0.1.0-beta.1 (January 1, 2026). See the workspace release notes for highlights and upgrade guidance.

Overview

sklears-cross-decomposition offers Partial Least Squares (PLS) regressors/classifiers, Canonical Correlation Analysis (CCA), and related cross-decomposition utilities. The APIs mirror scikit-learn 1.5, while Rust-native optimizations deliver consistent performance gains.

Key Features

• PLS Family: PLSRegression, PLSCanonical, PLSRegressionCV, and sparse extensions.
• CCA & Variants: Dense and sparse canonical correlation, along with GPU-accelerated solvers.
• Model Selection Integration: Works seamlessly with sklears pipelines, grid search, and feature engineering crates.
• Robust Numerics: Regularization, deflation strategies, and whitening controls ensure stability on real-world datasets.

Quick Start

use sklears_cross_decomposition::PLSRegression;
use scirs2_core::ndarray::{array, Array2};

let x: Array2<f64> = array![
    [0.0, 1.0, 2.0],
    [1.0, 2.0, 3.0],
    [2.0, 3.0, 4.0],
];
let y: Array2<f64> = array![
    [1.0, 0.0],
    [2.0, 1.0],
    [3.0, 2.0],
];

let pls = PLSRegression::builder()
    .n_components(2)
    .scale(true)
    .max_iter(500)
    .tol(1e-6)
    .build();

let fitted = pls.fit(&x, &y)?;
let y_pred = fitted.predict(&x)?;

Status

• Fully validated by the 11,292 passing workspace tests bundled with 0.1.0-beta.1.
• Benchmarks show 8–20× speedups versus scikit-learn for large PLS problems.
• Upcoming enhancements (incremental fit, streaming CCA) tracked in TODO.md.