sklears-inspection

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

Overview

sklears-inspection provides model interpretation tools, mirroring scikit-learn’s inspection module with additional Rust-first performance and visualization hooks.

Key Features

• Permutation Importance: CPU/GPU implementations with grouped feature support.
• Partial Dependence: Fast vectorized PDP and ICE computations for dense and sparse models.
• Feature Influence: SHAP-style approximations, ALE plots, and interaction strength metrics.
• Visualization Hooks: Data structures aligned with sklears-python plotting adapters.

Quick Start

use sklears_inspection::permutation_importance;
use sklears_ensemble::RandomForestClassifier;

let model = RandomForestClassifier::builder()
    .n_estimators(500)
    .n_jobs(-1)
    .build()
    .fit(&x_train, &y_train)?;

let importance = permutation_importance(
    &model,
    &x_val,
    &y_val,
    None,
    10,
)?;

println!("Mean importance for feature 0: {}", importance.importances_mean[0]);

Status

• Extensively covered by workspace integration tests; all 11,160 tests passed for 0.1.0-beta.1.
• Cross-crate sanity checks ensure compatibility with pipelines, model selection, and visualization crates.
• Further enhancements (GPU ICE surfaces, streaming importance) tracked in TODO.md.