Thursday, June 11, 2009

Nmix Paragraph

The Nmix program, coded by Peter Green in 1997, is a program for Bayesian analysis of univariate normal mixtures, implementing the approach of Richardson and Green, Journal of the Royal Statistical Society, B, 59, 731-792 (1997). It utilizes a reversible jump Markov Chain Monte-Carlo method, which gives the program the ability to determine the composition of a mixture with an unknown number of components, assuming that they are normally distributed. As it runs it divides the data into various numbers of components, k, splitting or combining components based on calculated probabilities. At the end of the run-cycle it estimates the posterior probabilities for each k. These probabilities are a calculated likelihood for each k-component fit. It also determines the most likely means and variances for each of the components included in each k.

I'll add a To Do list to this post, since I don't have enough to write for a full post.
1) Write a code that generates a bunch of bootstrapped data files, writes them to a file, then Nmixes that file, then writes the relevant data to another file. Should be pretty complicated.
2) Decide if the Nmix fail on bootstrapping things is really that typical.
3) Think of a third thing for this To Do list.

No comments:

Post a Comment