sklears-feature-selection

Latest release: 0.2.0 (July 14, 2026). See the workspace release notes for highlights and upgrade guidance.

Overview

sklears-feature-selection brings the complete scikit-learn feature selection toolbox to Rust, including filter, wrapper, and embedded methods. The crate underpins AutoML workflows, feature pipelines, and inspection utilities across the sklears project.

Key Features

• Filter Methods: VarianceThreshold, mutual information, ANOVA F-tests, chi-square tests, and more.
• Wrapper Methods: RFE/RFECV, SequentialFeatureSelector, model-based selectors with parallel evaluation.
• Embedded Techniques: L1-based selection (Lasso/ElasticNet/Ridge selectors), tree-based importance, stability selection, and feature importance scoring.
• Streaming Support: Online and concept-drift-aware selectors (streaming module: OnlineFeatureSelector, ConceptDriftAwareSelector) for evolving data; heavy scoring workloads parallelize via the parallel module. There is no CUDA/WebGPU backend in this crate.

Quick Start

use sklears_feature_selection::SequentialFeatureSelector;
use sklears_linear::LogisticRegression;

let estimator = LogisticRegression::new().max_iter(200);

let selector = SequentialFeatureSelector::new(estimator)
    .n_features_to_select(5)
    .direction("forward")?
    .cv(5);

let fitted = selector.fit(&x_train, &y_train)?;
let x_selected = fitted.transform(&x_train)?;

Status

• Validated by 290 passing crate tests in 0.2.0.
• Supports >99% of scikit-learn’s feature selection API surface.
• Additional milestones (distributed scoring, SHAP-guided selection) tracked in this crate’s TODO.md.