glmmEP.control {glmmEP}

Controlling generalized linear mixed model fitting via expectation propagation

Description

Function for optional use in calls to glmmEP() to control convergence values and other specifications for expectation propagation-based fitting of generalized linear mixed models.

Usage

glmmEP.control(confLev=0.95,BFGSmaxit=500,BFGSreltol=1e-10,
               EPmaxit=100,EPreltol=1e-5,NMmaxit=100,NMreltol=1e-10,
               quiet=FALSE,preTransfData=TRUE)

Arguments

confLev

Confidence level of confidence intervals expressed as a proportion (i.e. a number between 0 and 1). The default is 0.95 corresponding to 95% confidence intervals.

BFGSmaxit

Positive integer specifying the maximum number of iterations in the Broyden-Fletcher-Goldfarb-Shanno optimization phase. The default is 500.

BFGSreltol

Positive number specifying the relative tolerance value as defined in the R function optim() in the Broyden-Fletcher-Goldfarb-Shanno optimization phase. The default is 1e-10.

EPmaxit

Positive integer specifying the maximum number of iterations in the expectation propagation message passing iterations. The default is 100.

EPreltol

Positive number specifying the relative tolerance value for the expectation propagation message passing iterations. The default is 1e-5.

NMmaxit

Positive integer specifying the maximum number of iterations in the Nelder-Mead optimization phase. The default is 100.

NMreltol

Positive number specifying the relative tolerance value as defined in the R function optim() in the Nelder-Mead optimization phase. The default is 1e-10.

quiet

Flag for specifying whether or not glmmEP() runs ‘quietly’ or with progress reports printed to the screen. The default is FALSE.

preTransfData

Flag for specifying whether or not the predictor data are pre-transformed to the unit interval for fitting, with estimates, predictions and confidence intervals transformed to match the units of the original data before. The default is TRUE.

Value

A list containing values of each of the fifteen control parameters, packaged to supply the control argument to glmmEP. The values for glmmEP.control can be specified in the call to glmmEP.

Author(s)

Matt [e-mail] and James [e-mail]

References

Hall, P., Johnstone, I.M., Ormerod, J.T., Wand, M.P. and Yu, J.C.F. (2018). Fast and accurate binary response mixed model analysis via expectation propagation. Submitted.

Examples

library(glmmEP)

# Obtain simulated data corresponding to the simulation study in Section 4.1.2. of 
# Hall et al. (2018):

dataObj <- glmmSimData(seed=54321)
y <- dataObj$y  
Xfixed <- dataObj$Xfixed
Xrandom <- dataObj$Xrandom  
idNum <- dataObj$idNum

# Obtain and summarise probit mixed model fit with user control
# of some of the parameters in glmmEP.control():

myNMmaxit <- 500 ; myBFGSreltol <- 0.001

fitSimData <- glmmEP(y,Xfixed,Xrandom,idNum,
              control=glmmEP.control(NMmaxit=myNMmaxit,BFGSreltol=myBFGSreltol,quiet=TRUE))
summary(fitSimData)