HDMaxShrink: test-guided high-dimensional shrinkage

The package supports an independently learned CPSS core, a fresh exact-null SVD submodel on that core, and an unrestricted all-predictor dual-Ridge full endpoint. It also retains the all-predictor square-root-LASSO framework. Both center and RMS-scale X and y and fix the standardized intercept at zero. A conditional-Gaussian maximum partial-t test calibrates preliminary-test, Stein-type, positive-part, and separately reported pretest-protected positive-part estimators. Economy-SVD and compiled RcppArmadillo kernels avoid OLS Gram inverses, q by q covariance inverses, and p by p inverses. Earlier Ridge–MCP and restriction-projection modes remain available for sensitivity analysis and reproducibility.

Author

Maintainer: Bahadir Yuzbasi b.yzb@hotmail.com