dist_learn: Distributed learning for a longitudinal continuous-time...

In ziphsmm: Zero-Inflated Poisson Hidden (Semi-)Markov Models

Description

Distributed learning for a longitudinal continuous-time zero-inflated Poisson hidden Markov model, where zero-inflation only happens in State 1. Assume that priors, transition rates and state-dependent parameters can be subject-specific, clustered by group, or common. But at least one set of the parameters have to be common across all subjects.

Usage

dist_learn(ylist, timelist, prior_init, tpm_init, emit_init, zero_init,
  yceil = NULL, rho = 1, priorclust = NULL, tpmclust = NULL,
  emitclust = NULL, zeroclust = NULL, group, maxit = 100, tol = 1e-04,
  ncores = 1, method = "Nelder-Mead", print = TRUE, libpath = NULL, ...)

Arguments

ylist	list of observed time series values for each subject
timelist	list of time indices
prior_init	a vector of initial values for prior probability for each state
tpm_init	a matrix of initial values for transition rate matrix
emit_init	a vector of initial values for the means for each poisson distribution
zero_init	a scalar initial value for the structural zero proportion
yceil	a scalar defining the ceiling of y, above which the values will be truncated. Default to NULL.
rho	tuning parameters in the distributed learning algorithm. Default to 1.
priorclust	a vector to specify the grouping for state prior. Default to NULL, which means no grouping.
tpmclust	a vector to specify the grouping for state transition rates. Default to NULL, which means no grouping.
emitclust	a vector to specify the grouping for Poisson means. Default to NULL, which means no grouping.
zeroclust	a vector to specify the grouping for structural zero proportions. Default to NULL, which means no grouping.
group	a list containing group information.
maxit	maximum number iteration. Default to 100.
tol	tolerance in the terms of the relative change in the norm of the common coefficients. Default to 1e-4.
ncores	number of cores to be used for parallel programming. Default to 1.
method	method for the distributed optimization in the ADMM framework.
print	whether to print each iteration. Default to TRUE.
libpath	path for the ziphsmm library if not the default set up. Default to NULL.
...	Further arguments passed on to the optimization methods

Value

the maximum likelihood estimates of the zero-inflated hidden Markov model

References

Boyd, S., Parikh, N., Chu, E., Peleato, B. and Eckstein, J., 2011. Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends in Machine Learning, 3(1), pp.1-122.

Examples

## Not run: 
set.seed(930518)
nsubj <- 10
ns <- 5040
ylist <- vector(mode="list",length=nsubj)
timelist <- vector(mode="list",length=nsubj)
prior1 <- c(0.5,0.2 ,0.3 )
omega1 <- matrix(c(-0.3,0.2,0.1,
                  0.1,-0.2,0.1,
                  0.15,0.2,-0.35),3,3,byrow=TRUE)
prior2 <- c(0.3,0.3 ,0.4 )
omega2 <- matrix(c(-0.5,0.25,0.25,
                   0.2,-0.4,0.2,
                   0.15,0.3,-0.45),3,3,byrow=TRUE)
emit <- c(50,200,600)
zero <- c(0.2,0,0)
for(n in 1:nsubj){
 timeindex <- rep(1,ns)
 for(i in 2:ns) timeindex[i] <- timeindex[i-1] + sample(1:4,1)
 timelist[[n]] <- timeindex
 if(n<=5){
   result <- hmmsim.cont(ns, 3, prior1, omega1, emit, zero, timeindex)
   ylist[[n]] <- result$series
 }else{
   result <- hmmsim.cont(ns, 3, prior2, omega2, emit, zero, timeindex)
   ylist[[n]] <- result$series
 }
}
prior_init <- c(0.5,0.2,0.3)
emit_init <- c(50, 225, 650)
zero_init <- 0.2
tpm_init <- matrix(c(-0.3,0.2,0.1,0.1,-0.2,0.1,0.15,0.2,-0.35),3,3,byrow=TRUE)
M <- 3
priorclust <- NULL
tpmclust <- c(1,1,1,1,1,2,2,2,2,2)
zeroclust <- rep(1,10)
emitclust <- rep(1,10)
group <- vector(mode="list",length=2)
group[[1]] <- 1:5; group[[2]] <- 6:10
result <- dist_learn(ylist, timelist, prior_init, tpm_init, 
                    emit_init, zero_init,NULL, rho=1,priorclust,tpmclust,
                    emitclust,zeroclust,group,ncores=1,
                    maxit=50, tol=1e-4, method="CG", print=TRUE)

## End(Not run)

ziphsmm documentation built on May 2, 2019, 6:10 a.m.