ictonyx.explainers

SHAP-based feature importance across runs.

ictonyx.explainers.plot_shap_summary(model_wrapper, X_data, feature_names=None, plot_type='bar')[source]

Generates a SHAP summary plot for a model.

Parameters:

  • model_wrapper (BaseModelWrapper) – The trained model wrapper.

  • X_data (np.ndarray) – The data to explain.

  • feature_names (Optional[List[str]]) – List of feature names.

  • plot_type (str) – The type of plot to generate (“bar”, “dot”, “violin”).

ictonyx.explainers.plot_shap_waterfall(model_wrapper, X_data, sample_index=0, feature_names=None, class_index=0)[source]

Generates a SHAP waterfall plot for a single prediction.

Parameters:

  • model_wrapper (BaseModelWrapper) – The trained model wrapper.

  • X_data (np.ndarray) – The data to explain.

  • sample_index (int) – Index of the sample to explain.

  • feature_names (Optional[List[str]]) – List of feature names.

  • class_index (int) – For multi-class models, which class to explain.

ictonyx.explainers.plot_shap_dependence(model_wrapper, X_data, feature_name, feature_names=None, interaction_index=None, class_index=0)[source]

Generates a SHAP dependence plot showing how a feature affects predictions.

Parameters:

  • model_wrapper (BaseModelWrapper) – The trained model wrapper.

  • X_data (np.ndarray) – The data to explain.

  • feature_name (str) – Name of the feature to plot (or index if feature_names not provided).

  • feature_names (Optional[List[str]]) – List of feature names.

  • interaction_index (Optional[int]) – Feature index to use for coloring interaction effects.

  • class_index (int) – For multi-class models, which class to explain.

ictonyx.explainers.get_shap_feature_importance(model_wrapper, X_data, feature_names=None, class_index=0)[source]

Get feature importance scores based on mean absolute SHAP values.

Parameters:

  • model_wrapper (BaseModelWrapper) – The trained model wrapper.

  • X_data (np.ndarray) – The data to explain.

  • feature_names (Optional[List[str]]) – List of feature names.

  • class_index (int) – For multi-class models, which class to analyze.

Returns:

Mean absolute SHAP values for each feature.

Return type:

np.ndarray