EBICgolazo: Finds the optimal penalty parameter for the GOLAZO approach...

In pzwiernik/golazo: Regularized likelihood for Gaussian graphical models under a general asymmetric lasso-type penalty

Description

This function computes the EBIC criterion for a seried of rho parameters. The penalty matrices are rho L, rho U, where L and U are fixed in advance. If L, U are not specified, they will be set to the default value for the positive GOLAZO problem, that is, L_ij=0 and U_ij=1 for all i!=j.

Usage

EBICgolazo(
  S,
  n = NULL,
  L = NULL,
  U = NULL,
  tol = 1e-06,
  edge.tol = 1e-06,
  gamma = 0.5,
  rhomin = 0.01,
  rhomax = 1,
  nrhos = 10,
  verbose = TRUE
)

Arguments

S	a positive semidefinite matrix
n	the sample size
L	matrix of lower penalties (may contain -Inf)
U	matrix of upper penalties (may contain Inf)
tol	the convergence tolerance (default tol=1e-8) for the dual gap
edge.tol	the threshold for the normalized entries of K to be treated treated as zero
gamma	the EBIC parameter between 0 and 1 (default gamma=0.5)
rhomin	minimal rho to check
rhomax	maximal rho to check
nrhos	number of rhos in the interval (rhomin,rhomax)
verbose	if TRUE (default) the output will be printed.

Value

the list of rhos for which EBIC was computed

the list of teh corresponding EBIC values

the optimal rho

Examples

EBICgolazo(diag(10),n=10)

pzwiernik/golazo documentation built on Aug. 13, 2020, 4:15 p.m.