What is Kernel SHAP?
Kernel SHAP is a slow, perturbation-based Shapley approach that theoretically works for all types of models but is rarely used by teams in the wild (at least in production).
This explainability technique tends to be laborious in practice on anything but small data. It also tends to cause confusion among teams -- when teams complain about SHAP being slow, usually it’s because they tested Kernel SHAP.
Kernel SHAP works by calculating the contributions of each feature value to the prediction for an instance “x.” KernelSHAP is made up of five steps: examine coalitions, obtain a prediction for each coalition, calculate the weight for each using kernel SHAP, create a weighted linear model and return the Shapley values “k,” the linear model’s coefficients.