`bluecast.evaluation.shap_values`¶

Module to calculate SHAP values for a trained ML model.

The implementation is flexible and can be used for almost any ML model. The implementation is based on the SHAP library.

Module Contents¶

`shap_explanations`(→ Tuple[numpy.ndarray, shap.Explainer])	See explanations under:
`shap_waterfall_plot`(→ None)	Plot the SHAP waterfall plot.
`get_most_important_features_by_shap_values`(...)	Get the most important features by SHAP values.
`shap_dependence_plots`(→ None)	Plot the SHAP dependence plots for each column in the dataframe.

bluecast.evaluation.shap_values.shap_explanations(model, df: pandas.DataFrame) → Tuple[numpy.ndarray, shap.Explainer]¶: See explanations under: https://medium.com/rapids-ai/gpu-accelerated-shap-values-with-xgboost-1-3-and-rapids-587fad6822 :param model: Trained ML model :param df: Test data to predict on. :return: Shap values

bluecast.evaluation.shap_values.shap_waterfall_plot(explainer: shap.Explainer, indices: List[int], class_problem: Literal[binary, multiclass, regression]) → None¶: Plot the SHAP waterfall plot. :param explainer: SHAP Explainer instance :param indices: List of sample indices to plot. Each index represents a sample. :param class_problem: Class problem type (i.e.: binary, multiclass, regression) :return: None

bluecast.evaluation.shap_values.get_most_important_features_by_shap_values(shap_values: numpy.ndarray, df: pandas.DataFrame) → pandas.DataFrame¶: Get the most important features by SHAP values. :param shap_values: Numpy ndarray holding Shap values :param df: Pandas DataFrame :return: Pandas DataFrame with columns ‘col_name’ and ‘feature_importance_vals’

bluecast.evaluation.shap_values.shap_dependence_plots(shap_values: numpy.ndarray, df: pandas.DataFrame, show_dependence_plots_of_top_n_features: int) → None¶: Plot the SHAP dependence plots for each column in the dataframe. :param shap_values: Numpy ndarray holding Shap values :param df: Pandas DataFrame :param show_dependence_plots_of_top_n_features: Number of features to show the dependence plots for.