scparadise.scnoah.feature_importance_df

scparadise.scnoah.feature_importance_df#

scparadise.scnoah.feature_importance_df(explanations, output_index=None, absolute=True)[source]#

Create a feature-importance table from SHAP explanations.

Parameters:

  • explanations (shap.Explanation) – Object returned by explain() or another SHAP explainer.

  • output_index (int or None (default: None)) – Index of the model output to summarize if SHAP values contain multiple output dimensions. If None and exactly one output is present, that output is selected automatically. If multiple outputs are present, a ValueError is raised.

  • absolute (bool (default: True)) – If True, compute mean absolute SHAP value per feature. This is the standard global feature-importance measure. If False, compute mean signed SHAP value, which preserves direction of contribution.

Returns:

DataFrame sorted by descending importance with columns: - gene_name - gene_importance - mean_shap_value

Return type:

pandas.DataFrame