aflapan/sparseKOS: Implementation of sparse kernel optimal scoring

A package for non-linear binary classification with simultaneous sparse feature selection. This package implements an algorithm developed in [Lapanowski and Gaynanova, preprint] called sparse kernel optimal scoring (sparse KOS). It combines kernel discriminant analysis with sparse feature selection. The kernel discriminant analysis is done through a kernelized version of regularized optimal scoring, which is a regression-type problem. This package uses the Gaussian kernel. Sparse feature selection is accomplished by placing a weight lying in [-1,1] on each data feature, and then including a sparsity penalty on the weight vector. Sparse KOS alternates between solving the kernelized optimal scoring problem and minimzing a LASSO-type problem on a feature weight vector. This package also implements automatic kernel, ridge, and sparsity parameter selection methods.

install.packages("remotes")
remotes::install_github("aflapan/sparseKOS")