Gaussian Processes and Kernel Methods¶

A Gaussian-process mean with its uncertainty band

Models that are weighted sums of a product kernel, such as the RBF kernel, have exact Shapley values that are computed in polynomial time instead of over all coalitions. ProductKernelExplainer computes them for scikit-learn’s Gaussian-process regressors, support vector regressors, and binary support vector classifiers with an RBF kernel, and Explainer picks it for these models.

The examples explain Gaussian processes and kernel support vector machines. To benchmark approximators on these models, ProductKernelGame turns one into a game whose exact Shapley values are known.

Key API: ProductKernelExplainer · Explainer · ProductKernelGame

Exact Shapley Values of a Gaussian Process

Kernel SVMs: Exact Shapley Values for Many Features