|
Activity Number:
|
308
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #307869 |
|
Title:
|
Fully Nonparametric Bayesian Ensemble Modeling
|
|
Author(s):
|
Robert McCulloch*+ and Edward I. George and Hugh Chipman
|
|
Companies:
|
The University of Chicago Graduate School of Business and University of Pennsylvania and Acadia University
|
|
Address:
|
5807 S. Woodlawn Avenue, Chicago, IL, 60637,
|
|
Keywords:
|
data-mining ; markov chain monte carlo ; trees
|
|
Abstract:
|
Suppose we would like to learn the relationship between y and a high dimensional vector x based on a limited number of observations. In "BART: Bayesian Additive Regression Trees" (2006), Chipman, George and McCulloch develop a fully Bayesian approach for discovering and drawing inference about an unknown function f based only on assuming y = f(x) + e with iid normal errors. In the spirit of "ensemble models," BART approximates f by a sum of many simple regression tree models, each of which are kept small with a strong regularization prior. In this work, we further extend the flexibility of the BART approach by relaxing the simple iid normal error specification and replacing it with a Dirichlet process model for the errors. Various specification and prior choices are explored. The costs as well as the benefits of this more flexible approach are illustrated.
|
