kernelPSI

Different post-selection inference strategies for kernel selection, as described in "kernelPSI: a Post-Selection Inference Framework for Nonlinear Variable Selection", Slim et al., Proceedings of Machine Learning Research, 2019, . The strategies rest upon quadratic kernel association scores to measure the association between a given kernel and an outcome of interest. The inference step tests for the joint effect of the selected kernels on the outcome. A fast constrained sampling algorithm is proposed to derive empirical p-values for the test statistics.