Dear R users:
I was trying to fit a HMM with mixture of Gaussian into the dataset, and I
tried to implement it by R2Winbugs. But I got the following errer.
*
Error in FUN(X[[1L]], ...) :
.C(..): 'type' must be "real" for this format*
Does anybody know what's the problem? Does R2Winbugs accept some matrix as
inits? I would really appreciate your help. Thank you very much.
The attached are codes of R and Winbugs.
-------------------------------------------------------------------------------------------------------------------------
library(R2WinBUGS)
library(MCMCpack)
library(coda)
## the input of the dataset
X=read.csv("X.csv",header=FALSE)
X=as.matrix(X) # transform the data into matrix
## parameter setting
N=nrow(X) # # of servers
T=ncol(X) # Time
m=sum(X)/(N*T) # mean of the training set
M=matrix(m,nrow=N,ncol=T)
s=sum((X-M)^2)/(N*T) # std of the training set
K=3 # # of clusters
alpha=0.5 # parameter for Dirichlet distn
sigmae=0.5 # var of cluster mean mu
q1=rep(1/K,K) # prior for Z(n,1)
## MCMC sampling
data=list("X","m","s","N","T","K","alpha","sigmae","q1")
inits=function(){list(a0=rbeta(1,1,1),
qx=matrix(rgamma(K^2,alpha,1),nrow=K), sigma0.r=rbeta(K,1,1))}
model.sim=bugs(data,inits,model.file="model.txt",parameters=c("mu","sigma"),
n.chains=3,n.iter=3500,n.burnin=500,n.thin=1,bugs.directory="C:/Users/t-wec/Desktop/WinBUGS14",codaPkg=T,debug=T)
mcmcout=read.bugs(model.sim)
summary(mcmcout)
------------------------------------------------------------------------------------------------------------------------
model
{
# cluster parameters mu and tau
tau1 <- (1-a*a)*taue
taue <- 1/sigmae
for (k in 1:K)
{
# cluster mean mu
mu[k,1] ~ dnorm(m,tau1)
for (t in 2:T)
{
mu[k,t] ~ dnorm(meanmu[k,t],taue)
meanmu[k,t] <- m*(1-a)+a*mu[k,t-1]
}
# cluster varicance tau
sigma0.r[k] ~ dbeta(1,1)
sigma.r[k] <- s*sigma0.r[k]
sigma[k] <- sigma.r[k]*sigma.r[k]
tau[k] <- 1/sigma[k]
}
# cluster indicator Z and observation X
for (n in 1:N)
{
Z[n,1] ~ dcat(q1[1:K])
X[n,1] ~ dnorm(mu[Z[n,1],1],tau[Z[n,1]])
for (t in 2:T)
{
Z[n,t] ~ dcat(q[Z[n,t-1],1:K])
X[n,t] ~ dnorm(mu[Z[n,t],t],tau[Z[n,t]])
}
}
# prior on transition matrix Q
# each row of Q has a Dirichlet prior realized by Gamma
for (k in 1:K)
{
for (l in 1:K)
{
q[k,l] <- qx[k,l]/sum(qx[k,1:K])
qx[k,l] ~ dgamma(alpha,1)
}
}
# prior on regression coefficient: uniform on [-1,1]
a0 ~ dbeta(1,1)
a <- a0*2-1
}
Wei Chen
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
