glmmlassoControl: Options for the glmmlasso algorithm

Description Usage Arguments Details References

View source: R/glmmlassoControl.R

Description

Definition of various kinds of options in the algorithm.

Usage

1
2
3
4
5
6
7
glmmlassoControl(family, verbose = 0, maxIter = 200, number = 0,
CovOpt=c("nlminb"), fctSave = TRUE, a_init = 1, delta = 0.5,
rho = 0.1,gamm = 0, lower = 10^(-6),
upper = ifelse(family == "binomial", 10^5,10^3), seed = 418,
maxArmijo = 20, min.armijo = TRUE, thres = 10^(-4),
tol1 = 10^(-6), tol2 = 10^(-6), tol3 = 10^(-3), tol4 = 10^(-8),
gradTol = 10^(-3))

Arguments

family

a GLM family. Currently implemented are "binomial" (default) and "poisson".

verbose

integer. 0 prints no output, 1 prints the outer iteration step, 2 prints the current function value, 3 prints the values of the convergence criteria

maxIter

maximum number of (outer) iterations

number

integer. Determines the active set algorithm. The zero fixed-effects coefficients are only updated each number iteration. Use 0 ≤ number ≤ 10.

CovOpt

character string indicating which covariance parameter optimizer to use. Currently, only "nlminb" is implemented

fctSave

Should all evaluation of the objective function be stored? It may help to identify the convergence pattern of the algorithm.

a_init

α_{init} in the Armijo step.

delta

δ in the Armijo step.

rho

ρ in the Armijo step.

gamm

γ in the Armijo step.

lower

lower bound for the Hessian

upper

upper bound for the Hessian

seed

set.seed in order to choose the same starting value in the cross-validation for the fixed effects

maxArmijo

maximum number of steps to be chosen in the Armijo step. If the maximum is reached, the algorithm continues with optimizing the next coordinate.

min.armijo

logical. If TRUE, the smallest l in the Armijo step is increased, as suggested in Tseng and Yun (2009). Otherwise l always starts with 0.

thres

if a variance or covariance parameter has smaller absolute value than thres, the parameter is set to exactly zero,

tol1

convergence tolerance for the relative change in the function value

tol2

convergence tolerance for the relative change in the fixed-effects parameters

tol3

convergence tolerance for the relative change in the covariance parameters

tol4

convergence tolerance in the PIRLS algorithm

gradTol

the tolerance for the gradient accepted without giving a warning

Details

For the Armijo step parameters, see Bertsekas (2003).

References

Dimitri P. Bertsekas (2003) Nonlinear Programming, Athena Scientific.


glmmixedlasso documentation built on May 31, 2017, 3:34 a.m.