cv_covariance_matrices_block_descent: Projected covariance matrices for BD-CoCoLasso

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

Description

Creating projected nearest positive semi-definite covariance matrices for the cross validation step of the BD-CoCoLasso algorithm. In that case, the design matrix must be organized as follozs : uncorrupted features must be the first block of the matrix, and corrupted features must be the second block of the matrix.

Usage

cv_covariance_matrices_block_descent(
  K,
  mat,
  y,
  p,
  p1,
  p2,
  mu,
  tau = NULL,
  ratio_matrix = NULL,
  etol = 1e-04,
  noise = c("additive", "missing"),
  mode = "ADMM"
)

Arguments

K	Number of folds for the cross validation
mat	Covariance matrix to be projected
y	Response vector
p	Number of predictors
p1	Number of uncorrupted predictors
p2	Number of corrupted predictors
mu	Penalty parameter for the ADMM algorithm.
tau	Standard error of the additive error matrix when the chosen setting is the additive error setting
ratio_matrix	Observation matrix used in the missing data setting
etol	Tolerance used in the ADMM algorithm
noise	Type of setting chosen : additive or missing
mode	ADMM or HM

Value

list containing

• sigma_global projected matrix for mat

• rho_global rho parameter for mat

• list_matrices_lasso list of the projected matrices for mat deprived of the k-fold during cross validation

• list_matrices_error list of the projected matrices for the k-fold of mat

• list_rho_lasso list of the modified rho for mat deprived of the K-fold during cross validation

• list_rho_error list of the modified rho for the k-fold of mat

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