Sparse Gaussian Regression with the Lambert Penalty
lambertReg-package.RdAn R/C++ implementation of fixed-shape Lambert regression with regularization paths, five-fold cross-validation by default, and explicit numerical diagnostics.
Details
The package implements the main estimator in the arXiv preprint
The Lambert Penalty: Logarithmic Shrinkage for Sparse Regression
by Bahadir Yuzbasi (2026), arXiv:2610.09627 [stat.ME], available at
https://arxiv.org/abs/2610.09627. It does not implement the extended shape family,
Lambert-Min post-selection procedures, or comparator estimators.
Start with cv.lambert to select a penalty, then use coef,
predict and plot. Use lambert for a path with specified
penalties. All predictions and reported coefficients use the original input
scale. Training-only predictor centering and RMS scaling are automatic.
The worked guide is available through
vignette("introduction", package = "lambertReg").
Source code and issue reporting are available at
https://github.com/byuzbasi/lambertReg.