Explainability

Explainability

SHAP (SHapley Additive exPlanations)

Global and local feature importance for any sklearn-compatible model.

from scomp_link import ShapExplainer

explainer = ShapExplainer(model, X_train[:100])  # background data
explainer.explain(X_test)                        # compute SHAP values

# Global importance
importance = explainer.feature_importance()      # DataFrame: feature, mean_abs_shap
fig = explainer.plot_importance(top_n=15)        # Plotly bar chart

# Local explanations
explainer.plot_waterfall(idx=0)                  # single prediction
explainer.plot_beeswarm()                        # all predictions

LIME (Local Interpretable Model-agnostic Explanations)

Per-instance explanations.

from scomp_link import LimeExplainer

lime = LimeExplainer(model, X_train, task='regression')
explanation = lime.explain_instance(X_test.iloc[0], num_features=10)
fig = lime.plot_explanation(explanation)

# Aggregated importance
importance = lime.feature_importance(X_test, n_samples=50)

CLI

scomp-link explain --artifact model.scomp --data test.csv --n-samples 100 --output importance.csv