tabensemb.trainer.Trainer.cal_shap#

method

Trainer.cal_shap(program: str, model_name: str, **kwargs) → ndarray[source]#

Calculate SHAP values using a specified model. If the model base is a tabensemb.model.TorchModel, the shap.DeepExplainer is used. Otherwise, shap.KernelExplainer is called, which is much slower, and shap.kmeans is called to summarize the training data to 10 samples as the background data and 10 random samples in the testing set is explained, which will bias the results.

Parameters:

program: The selected model base.
model_name: The selected model in the model base.
kwargs: kwargs for tabensemb.model.AbstractModel.cal_shap()

Returns:

attr: The SHAP values. If the model base is a TorchModel, all features including derived unstacked features will be included. Otherwise, only all_feature_names() will be considered.