Efficient Approximation of Asymmetric Shapley Values Using Functional Decomposition

Abstract Asymmetric Shapley values (ASVs) are an extension of that allow a user to incorporate partial causal knowledge into the explanation process. Unfortunately, computing ASVs requires sampling permutations, which quickly becomes computationally expensive. We propose A-PDD-SHAP, algorithm employs functional decomposition approach approximate at speed orders magnitude faster compared permutation sampling, significantly reduces amortized complexity when many explanations needed. Apart from this, once A-PDD-SHAP model is trained, it can be used compute both symmetric and asymmetric without having re-train or re-sample, allowing for very efficient comparisons between different types explanations.