glmmLassoControl: Control Values for 'glmmLasso' fit

In glmmLasso: Variable Selection for Generalized Linear Mixed Models by L1-Penalized Estimation

Control Values for glmmLasso fit

Description

The values supplied in the function call replace the defaults and a list with all possible arguments is returned. The returned list is used as the control argument to the glmmLasso function.

Usage

glmmLassoControl(nue=1,index=NULL,smooth=NULL, start=NULL, q_start=NULL, 
                 center = TRUE, standardize = TRUE, steps=1000, 
                 method="EM", overdispersion=FALSE,     
                 epsilon=1e-4, maxIter=200, print.iter=FALSE, 
                 print.iter.final=FALSE, method.final="EM", 
                 eps.final=1e-4, Q.fac=5, complexity="hat.matrix",...)

Arguments

nue	weakness of the learner. Choose 0 < nue =< 1. Default is 1.
index	vector which defines the grouping of the variables. Components sharing the same number build a group and factor variables get a single number (and are automatically treated as a group). Non-penalized coefficients are marked with NA.
smooth	a list specifying the formula of the smooth terms, together with the number of basis functions nbasis, the degree of the B-splines spline.degree, the order of differences that is used for penalization diff.ord and finally a correspodning penalty parameter penal.
start	a vector containing starting values for fixed and random effects of suitable length. Default is a vector full of zeros.
q_start	a scalar or matrix of suitable dimension, specifying starting values for the random-effects variance-covariance matrix. Default is a scalar 0.1 or diagonal matrix with 0.1 in the diagonal, depending on the dimension of the random effects.
center	logical. If true, the columns of the design matrix will be centered (except a possible intercept column).
standardize	logical. If true, the design matrix will be blockwise orthonormalized such that for each block X^TX = n 1 (*after* possible centering).
steps	the number of iterations. Default is 1000.
method	two methods for the computation of the random-effects variance-covariance parameter estimates can be chosen, an EM-type estimate and an REML-type estimate. The REML-type estimate uses the nlminb or the bobyqa function for optimization, depending on the dimension of the random effects. Default is EM.
overdispersion	logical scalar. If FALSE, no scale parameter is derived, if TRUE, in each iteration a scale parameter is estimated by use of Pearson residuals. This can be used e.g. to fit overdispersed Poisson models. Default is FALSE. If the Gaussian family is used, overdispersion is automatically set TRUE.
epsilon	controls the speed of convergence. Default is 1e-4.
maxIter	the number of iterations for the final Fisher scoring re-estimation procedure. Default is 200.
print.iter	logical. Should the number of iterations be printed? Default is FALSE.
print.iter.final	logical. Should the number of iterations in the final re-estimation step be printed? Default is FALSE.
method.final	two methods for the computation of the random-effects variance-covariance parameter estimates for the final Fisher scoring re-estimation procedure can be chosen, an EM-type estimate and an REML-type estimate. The REML-type estimate uses the bobyqa function for optimization. Default is EM.
eps.final	controls the speed of convergence in the final re-estimation. Default is 1e-4.
Q.fac	Factor which controls the interval on which is searched for the optimal parameters of the random-effects variance-covariance matrix, if method.final="REML". Default is 5.
complexity	Character which determines how the model complexity is computed. Default is "hat.matrix", which sums up the trace of the corresponding hat matrix. Alternatively, simply the number of estimated (non-zero) parameters can be used by setting complexity="non-zero".
...	Futher arguments to be passed.

Value

a list with components for each of the possible arguments.

Author(s)

Andreas Groll [e-mail]

See Also

glmmLasso, bobyqa

Examples

# Use REML estimates for random effects covariance parameters
# and lighten the convergence criterion 
glmmLassoControl(method="REML", epsilon=1e-4)

glmmLasso documentation built on Aug. 23, 2023, 5:06 p.m.