pmhtutorial (version 1.0.0)

example2_lgss: Parameter estimation in a linear Gaussian state space model

Description

Minimal working example of parameter estimation in a linear Gaussian state space model using the particle Metropolis-Hastings algorithm with a fully-adapted particle filter for providing an unbiased estimator of the likelihood. The code estimates the parameter posterior for one parameter using simulated data.

Usage

example2_lgss(nIter = 5000)

Arguments

nIter

The number of iterations in the PMH algorithm. 100 iterations takes about ten seconds on a laptop to execute. 5000 iterations are used in the reference below. The length of the burn-in is calculated as one fifth of nIter.

Value

Returns the estimate of the posterior mean.

Details

The Particle Metropolis-Hastings (PMH) algorithm makes use of a Gaussian random walk as the proposal for the parameter. The PMH algorithm is run using different step lengths in the proposal. This is done to illustrate the difficulty when tuning the proposal and the impact of a too small/large step length.

References

Dahlin, J. & Schoen, T. B. "Getting started with particle Metropolis-Hastings for inference in nonlinear dynamical models." pre-print, arXiv:1511.01707, 2015.

Examples

# NOT RUN {
# Use a short run when testing the code on CRAN
# Set nIter manually if running on local computer
# nIter=1000 is usually enough
if(nchar(Sys.getenv("LONG_TEST")) != 0) {
  nIter=7500
} else {
  nIter=100
}
 
# Run the example script
example2_lgss(nIter=100) 
# }