Forums

Gibbs sampling jump/step size

Started by Benjamin S. March 22, 2012
I'm trying to perform Gibbs sampling. I've written the algorithm but I'm 
not sure how to calculate the jump/step size so that successful 
iterations would be independent (Gibbs is MCMC).

I have the data using a step of 1 and I've calculated the autocorrelation 
of the data. So I thought the jump size is the size when the 
autocorrelation crosses the x axis but this value seems to depend on how 
many iterations you perform.

BTW by step size I mean that you take every nth sample (skipping the n-1 
samples in between) so that successful data terms are independent.