SMPracticals (version 1.4-3.1)

hus.gibbs: Gibbs Sampler for Poisson Changepoint Model, Practical 11.6

Description

This function implements a Gibbs sampler for the Poisson changepoint model applied to the HUS data used in Example 4.40 and Practical 11.6 of Davison (2003), which should be consulted for details.

Usage

hus.gibbs(init, y, R = 10, a1 = 1, a2 = 1, c = 0.01, d = 0.01)

Value

A matrix of size R x 7, whose first five columns contain the values of the parameters for the iterations. Columns 6 and 7 contain the log likelihood and log prior for that iteration.

Arguments

init

Initial values for parameters

y

A series of Poisson counts

R

Number of iterations of sampler

a1

Value of a hyperparameter

a2

Value of a hyperparameter

c

Value of a hyperparameter

d

Value of a hyperparameter

Author

Anthony Davison ([e-mail])

Details

This is provided simply so that readers spend less time typing. It is not intended to be robust and general code.

References

Davison, A. C. (2003) Statistical Models. Cambridge University Press. Practical 11.6.

Examples

## From Example 11.6:
hus <- c(1,5,3,2,2,1,0,0,2,1,1,7,11,4,7,10,16,16,9,15)
system.time( gibbs.out <- hus.gibbs(c(5, 5, 1, 1, 2), hus, R=1000))
plot.ts(gibbs.out[,1], main="lambda1") # time series plot for lam1
plot.ts(gibbs.out[,2], main="lambda1") # time series plot for lam2
plot.ts(gibbs.out[,6], main="log lik") # and of log likelihood
table(gibbs.out[,5])  # tabulate observed values of tau
rm(hus)