CatReg: Solution Paths for Linear and Logistic Regression Models with SCOPE Penalty

Benjamin Stokell and Rajen Shah and Grose, Daniel (2020) CatReg: Solution Paths for Linear and Logistic Regression Models with SCOPE Penalty. UNSPECIFIED.

Abstract

Computes solutions for linear and logistic regression models with a nonconvex penalty (SCOPE) in an efficient path-wise fashion (Stokell, Shah and Tibshirani 2020, <arXiv:2002.12606>). The scaling of the solution paths is selected automatically. Includes functionality for selecting tuning parameter lambda by k-fold cross-validation and early termination based on information criteria. Solutions are computed by cyclical block-coordinate descent, iterating an innovative dynamic programming algorithm to compute exact solutions for each block.