Sparse Learning with Convex and Concave Penalties
Description
Fast regularization paths for sparse Gaussian, binomial, Poisson,
square-root-lasso, and multinomial models with lasso, smoothly clipped
absolute deviation, or minimax concave penalties. Computation uses
pathwise coordinate optimization, active-set updates, warm starts,
screening rules, Proximal Newton iterations, quadratic majorization, and
adaptive local linear approximation where appropriate. Core solvers are
implemented in C++, and coefficient paths are returned as
Matrix-compatible objects.