Abstract:
|
With hundreds of beer styles and thousands of breweries in hundreds of countries, a model that describes the flavour preferences for beer inherently has a large number of parameters. Using a hierarchical Bayesian model for spatial-flavour preference clusters based on text beer reviews as a high dimensional application, we present a technique for high dimensional MCMC diagnostics. Standard diagnostic tools such as Gelman-Rubin, Geweke, and other MCMC Diagnostics were designed for assessing convergence one parameter at a time. The proposed approach uses multiple MCMC chains and provides a test for convergence by using a classifier based on General Additive Models.
|