tabensemb.trainer.Trainer.cal_feature_importance#

method

Trainer.cal_feature_importance(program: str, model_name: str, method: str = 'permutation', **kwargs) → Tuple[ndarray, List[str]][source]#

Calculate feature importance using a specified model. If the model base is a tabensemb.model.TorchModel, captum or shap is called to make permutations. If the model base is only a tabensemb.model.AbstractModel, the calculation will be much slower.

Parameters:

program: The selected model base.
model_name: The selected model in the model base.
method: The method to calculate importance. “permutation” or “shap”.
kwargs: kwargs for tabensemb.model.AbstractModel.cal_feature_importance()

Returns:

attr: Values of feature importance.
importance_names: Corresponding feature names. If the model base is a TorchModel, all features including derived unstacked features will be included. Otherwise, only all_feature_names() will be considered.