Implementation of Bayesian Nonparametric Inference Based on Beta Processes
Duke University, Imperial College, Kimberly-Clark Corp., Medical College of Wisconsin
Nov 30 1993
Hjort (1990) constructs prior distributions for cumulative hazard rates using stochastic processes with nonnegative independent increments. A particular class of processes termed beta processes is introduced there and is shown to constitute a conjugate class. In this paper we develop and algorithm that enable approximate random variate generation from the posterior, given the Levy formula for the Laplace transform of the process. Bayesian nonparametric inference using failure time data is illustrated.
Keywords:
Beta process, Levy process, Hazard functions, Mixture densities, non-parametric inference, fixed points of discontinuityManuscript:
