pirls: Create an approximate deviance evaluation function for GLMMs...

Description Usage Arguments Details Value

View source: R/pirls.R

Description

A pure R implementation of the penalized iteratively reweighted least squares (PIRLS) algorithm for computing generalized linear mixed model deviances. The purpose is to clarify how PIRLS works without having to read through C++ code, and as a sandbox for trying out modified versions of PIRLS.

Usage

1
2
3
pirls(X, y, Zt, Lambdat, thfun, theta, weights, offset = numeric(n),
  eta = numeric(n), family = binomial, tol = 10^-6, npirls = 30,
  nstephalf = 10, nAGQ = 1, verbose = 0L, ...)

Arguments

glmod

output of glFormula

y

response

eta

linear predictor

family

a glm family object

weights

prior weights

offset

offset

tol

convergence tolerance

npirls

maximum number of iterations

nAGQ

either 0 (PIRLS for u and beta) or 1 (u only). currently no quadature is available

verbose

verbose

Details

pirls1 is a convenience function for optimizing pirls under nAGQ = 1. In particular, it wraps theta and beta into a single argument thetabeta.

Value

A function for evaluating the GLMM Laplace approximated deviance


lme4/lme4pureR documentation built on May 21, 2019, 7:34 a.m.