Real NamePhilipp Neubauer
I am a post-doctoral research associate at Rutgers University, working in fish and fisheries ecology/oceanography.
by Philipp, last updated 2/8/12, sharing set to PublicGeochemical tags are often employed to classify fish in a sample of mixed origin (stocks) to distinct spawning habitats. In order for these individuals to be uniquely attributed to one of these sources, this approach generally assumes that all potential sources have been characterized in terms of their geochemical signature. This assumption is rarely justified in marine environments, and statistical methods that can accommodate this problem are therefore essential. This work develops non-parametric Bayesian mixture models for isotopic and geochemical signatures that directly infer the most likely number of distinguishable sources represented in a mixed sample, both in the absence and presence of a baseline sample. A marginal clustering framework is used to illustrate the degree of relatedness of mixed sample fish in terms of geographical sources.
The functions below rely on the lightspeed toolbox for matlab, which can be found at http://research.microsoft.com/en-us/um/people/minka/software/lightspeed/ - please install this according to instructions before trying to run the sampler.
The Toy Data Creator produces a number of simulated examples to get a feel for the model. I highly recommend taking the time to go through that file, changing the parameters and running examples to see what this method can and cannot do...