lasso_covariance: Lasso in covariance form

In celiaescribe/BDcocolasso: Implementation of CoCoLasso and Block Descent CoCoLasso

Description

Solve the least squares loss with lasso penalty written in a form with the covariance matrix : \frac{1}{2} β^{'} Σ β - ρ^{'} β + λ \|β\|_1

Usage

lasso_covariance(
  n,
  p,
  lambda,
  control = list(maxIter = 1000, optTol = 10^(-5), zeroThreshold = 10^(-6)),
  XX,
  Xy,
  beta.start,
  penalty = c("lasso", "SCAD")
)

Arguments

n	Number of samples of the design matrix
p	Number of features of the matrix
lambda	penalty parameter
control	Including control parameters : max of iterations, tolerance for the convergence of the error, zero threshold to put to zero small beta coefficients
XX	Design matrix corresponding to \frac{1}{n} X'X or a modified version in the case of CoCoLasso
Xy	Rho parameter corresponding to \frac{1}{n} X'y or a modified version in the case of CoCoLasso
beta.start	Initial value of beta
penalty	Type of penalty used : can be lasso penalty or SCAD penalty

Value

list containing

• coefficients : Coefficients corresponding to final beta after convergence of the algoritm

• coef.list : Matrix of coefficients for beta for all iterations

• num.it Number of iterations

celiaescribe/BDcocolasso documentation built on Feb. 11, 2020, 11:41 p.m.