lava (version 1.6)

mixture: Estimate mixture latent variable model.

Description

Estimate mixture latent variable model

Usage

mixture(x, data, k = length(x), control = list(), vcov = "observed", ...)

Arguments

x

List of lvm objects. If only a single lvm object is given, then a k-mixture of this model is fitted (free parameters varying between mixture components).

data

data.frame

k

Number of mixture components

control

Optimization parameters (see details) # @param type Type of EM algorithm (standard, classification, stochastic)

vcov

of asymptotic covariance matrix (NULL to omit)

...

Additional arguments parsed to lower-level functions

Details

Estimate parameters in a mixture of latent variable models via the EM algorithm.

The performance of the EM algorithm can be tuned via the control argument, a list where a subset of the following members can be altered:

start

Optional starting values

nstart

Evaluate nstart different starting values and run the EM-algorithm on the parameters with largest likelihood

tol

Convergence tolerance of the EM-algorithm. The algorithm is stopped when the absolute change in likelihood and parameter (2-norm) between successive iterations is less than tol

iter.max

Maximum number of iterations of the EM-algorithm

gamma

Scale-down (i.e. number between 0 and 1) of the step-size of the Newton-Raphson algorithm in the M-step

trace

Trace information on the EM-algorithm is printed on every traceth iteration

Note that the algorithm can be aborted any time (C-c) and still be saved (via on.exit call).

See Also

mvnmix

Examples

# NOT RUN {
# }
# NOT RUN {
m0 <- lvm(list(y~x+z,x~z))
distribution(m0,~z) <- binomial.lvm()
d <- sim(m0,2000,p=c("y<-z"=2,"y<-x"=1),seed=1)

## unmeasured confounder example
m <- baptize(lvm(y~x));
covariance(m,~x) <- "v"
intercept(m,~x+y) <- NA

set.seed(42)
M <- mixture(m,k=2,data=d,control=list(trace=1,tol=1e-6))
summary(M)
lm(y~x,d)
estimate(M,"y~x")
## True slope := 1
# }
# NOT RUN {
# }