Fit a square-root lasso model with an RcppArmadillo solver — sqrt_lasso_hd • HDMaxShrink

This lower-level function fits an unconstrained square-root lasso after centering and scaling the design. It never forms or inverts a Gram matrix.

Usage

sqrt_lasso_hd(
  X,
  y,
  lambda = NULL,
  penalty_factor = NULL,
  max_iter = 30000L,
  tol = 5e-06,
  check_every = 50L,
  verbose = FALSE
)

Arguments

X: Numeric design matrix.
y: Numeric response vector.
lambda: Positive penalty. The default is 1.1 * sqrt(2 * log(2 * p) / n).
penalty_factor: Nonnegative penalty multiplier for every coefficient.
max_iter: Maximum primal-dual iterations.
tol: Convergence tolerance.
check_every: Iteration interval for KKT checks.
verbose: Print periodic solver diagnostics.

Value

A list containing coefficients, intercept, and solver diagnostics.