StEM: Stochastic EM algorithm

In kkholst/lava.nlin: Non-linear latent variable models via StEM and Laplace approximation

Description

Stochastic EM algorithm

Usage

StEM(model, modelpar, theta0, data, cluster = 1:nrow(data), eta,
  control = list(), Mfun, CondVarEta, update = FALSE, m = 1, nsim = 50,
  iter = 50, stepsize = 1, burnin = nsim - 1, plot = FALSE,
  idx = 1:length(theta0), printidx = idx, printvar = FALSE, ...)

Arguments

model	String defining model
modelpar	Model parameters
theta0	Initial parameter vector
data	data-frame
cluster	See Estep
eta	See Estep
control	See Estep
Mfun	Function defining M-step
CondVarEta	Variance of proposal distribution (defaults to the identity matrix).
update	Adaptive algorithm, where the variance of the proposal distribution is updated in each E-step from the empirical distribution of the latent variables given the data (obtained from previous E-step).
m	The number of imputations to base the E-step on (m=1 => StEM)
nsim	Number of samples to draw in the E-step (a burnin period is needed!)
iter	Number of iterations of the EM-algorithm
stepsize	Scaling of the variance of the proposal distribution
burnin	For development testing only
plot	Plot coordinates 'idx' of chain
idx	Trace these coordinates
printidx	Print these coordinates in each M-step
printvar	Boolean indicating whether variance parameters should be printed
...	Additional parameters parsed on to lower-level functions.

Value

list

Author(s)

Klaus K. Holst

Examples

## Not run: 
modelpar <- list(nlatent=3, ny1=4, ny2=3, npred=2)
modelpar$model <- "nsem1"
nparreg <- with(modelpar, ny1+ny2 + (ny1-1) + (ny2-1) + 2 + npred)
nparvar <- with(modelpar, ny1+ny2+nlatent)
modelpar$theta <- rep(5,nparreg+nparvar)
npar <- with(modelpar, ny1+ny2 + (ny1-1) + (ny2-1) + npred + 2 + ny1 +
ny2 + nlatent)
aa <- StEM(modelpar$model,modelpar=modelpar,theta0=modelpar$theta,data=yy,cluster=1:NROW(yy),nsim=200,iter=500,plot=FALSE,update=FALSE,stepsize=0.2,idx=15:16,m=5)
plot(aa,idx=15:16)
bb <- lava.nlin::restart(aa,stepsize=0.1,nsim=1000,update=TRUE,m=1)
bb

## End(Not run)

kkholst/lava.nlin documentation built on May 20, 2019, 10:47 a.m.