hmhmm {bayess}

Estimation of a hidden Markov model with 2 hidden and 4 observed states

Description

This function implements a Metropolis within Gibbs algorithm that produces a sample on the parameters p_{ij} and q^i_j of the hidden Markov model (Chapter 7). It includes a function likej that computes the likelihood of the times series using a forward-backward algorithm.

Usage

hmhmm(M = 100, y)

Arguments

M

Number of Gibbs iterations

y

times series to be modelled by a hidden Markov model

Details

The Metropolis-within-Gibbs step involves Dirichlet proposals with a random choice of the scale between 1 and 1e5.

Value

BigR

matrix of the iterated values returned by the MCMC algorithm containing p_{11} and p_{22}, transition probabilities, and q^1 and q^2, vector of probabilities for both latent states

olike

sequence of the log-likelihoods produced by the MCMC sequence

Examples

res=hmhmm(M=500,y=sample(1:4,10,rep=TRUE))
plot(res$olike,type="l",main="log-likelihood",xlab="iterations",ylab="")