RKHS-SHAP: Shapley Values for Kernel Methods

Authors: Siu Lun Chau, Robert Hu, Javier González, Dino Sejdinovic

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate specific properties of RKHS-SHAP and Shapley regularisers using four synthetic experiments, because these properties are best illustrated under a fully controlled environment.
Researcher Affiliation Collaboration Siu Lun Chau Department of Statistics University of Oxford Robert Hu Amazon London Javier Gonzalez Microsoft Research Cambridge Dino Sejdinovic School of Computer and Mathematical Sciences University of Adelaide
Pseudocode No The paper describes algorithms and mathematical formulations, but it does not include a distinct pseudocode block or algorithm listing.
Open Source Code Yes All code and implementations are made publicly available [3].
Open Datasets Yes We consider the following 2d Banana distribution B(b 1, v) from Sejdinovic et al. [34]: Sample Z N(0, diag(v, 1)) and transform the data by setting X1 = Z1 and X2 = b 1(Z2 1 v) + Z2. Regression labels are obtained from ftruth(X) = b 1(X2 1 v) + X2.
Dataset Splits Yes We use 70% of our data for training and 30% for testing. Lengthscales of the kernel are selected using median heuristic [12] and regularisation parameters are selected using cross-validation.
Hardware Specification No The paper does not specify the exact GPU/CPU models, memory, or other detailed hardware specifications used for experiments.
Software Dependencies No The paper mentions using the 'Python package shap [23]' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Lengthscales of the kernel are selected using median heuristic [12] and regularisation parameters are selected using cross-validation. Further implementation details and real world data illustrations are included in Appendix E.