Check the SHAP feature importance for a subset of features

Question

I`m trying to check the feature importance of an model I build using SHAP.

It works fine using the following, but I would like to have a graph with only a subset of features. Could this be possible?

shap.summary_plot(shap_values, features=X_train2, feature_names=X_train2.columns, max_display=30)

I tried to define the subset of features and feature_names using iloc on the dataset, like features=X_train.iloc[;23:], but it is not working.

Answer 1

You will need to subset both the features and the shap_values, otherwise they won't line up. For what it is worth the new SHAP API that uses shap.Explanation objects will make this easier since it allows the for the simple parallel slicing of features and SHAP values at the same time.