How to get the exact SHAP importance values?

[wishforgood]

For example I have a BoostRFE object, and have called the model.fit() function of it, then how can I get the exact list of SHAP importance corresponding to each features selected by the model's estimator (just like what shap.plot.bar shows)? The values of the output from shap.Explainer seems not that straightforward to me, I have not found any info about the shap or meaning of those values.

Or can I get the SHAP value directly through the BoostRFE object?

[wishforgood]

Oh maybe I should ask it in the git of shap

[wishforgood]

When I use the model.estimator_ to calculate SHAP values, I found that some features corresponds to NaN values, maybe because I used the min_features_to_select option? How can I get the SHAP values used to select the features?

[cerlymarco]

Hi,
in shap-hypetune, shap values are computed using shap.TreeExplainer with feature_perturbation="tree_path_dependent". shap importances are aggregated featurewise using this aggregation: np.abs(shap_values).mean(0). No method is available in shap-hypetune to extract/plot shap importances (since you can directly use the shap library).
To get the shap importances of the selected features, compute the shape values on your data (train or evaluation set using only the selected features) and aggregate them with the calculation np.abs(shap_values).mean(0)

shap-hypetune/shaphypetune/utils.py

Lines 36 to 44 in 47316d3

	explainer = TreeExplainer(
	model, feature_perturbation="tree_path_dependent")
	coefs = explainer.shap_values(X)
	if isinstance(coefs, list):
	coefs = list(map(lambda x: np.abs(x).mean(0), coefs))
	coefs = np.sum(coefs, axis=0)
	else:
	coefs = np.abs(coefs).mean(0)

if u support the project, don't forget to leave a star ;-)