I’ve seen Jennifer Hill and Ed George give great talks on Bayesian additive regression trees. It looked awesome. So why haven’t these papers appeared anywhere? All I can find are preprints.
Whassup with Bart?
I’ve seen Jennifer Hill and Ed George give great talks on Bayesian additive regression trees. It looked awesome. So why haven’t these papers appeared anywhere? All I can find are preprints.
Been wondering about the myself. But Robert Gramacy's extension to Treed Gaussian Processes is forthcoming in JASA.
One of the top papers on your list — about 4th or 5th down — appeared in JMR a couple of years ago.
A Direct Approach to Data Fusion – all 16 versions »
ZVI GILULA, RE MCCULLOCH, PE ROSSI – papers.ssrn.com
Page 1. A Direct Approach to Data Fusion Zvi Gilula Department of Statistics
Hebrew University Robert E. McCulloch Peter E. Rossi
appears in
Journal of Marketing Research, vol 43, Feb 2006
It's a marketing article, and JMR is a top journal in this field.
The paper below, third on your list, was published in the Journal of Marketing Research (a top marketing journal) in February, 2006.
A Direct Approach to Data Fusion – all 16 versions »
ZVI GILULA, RE MCCULLOCH, PE ROSSI – papers.ssrn.com
Page 1. A Direct Approach to Data Fusion Zvi Gilula Department of Statistics
Hebrew University Robert E. McCulloch Peter E. Rossi …
Maybe is the language they use. It is only intelligible to motivated Bayesians. A more accessible version is needed than this:
"Effectively, BART is a nonparametric Bayesian regression approach which uses dimensionally adaptive random basis elements. Motivated by ensemble methods in general, and boosting algorithms in particular, BART is defined by a statistical model: a prior and a likelihood."
Marketing of one's research is very important, they ain't doing themselves any favors.
I find a piece of research to be really good when it can be very impressive in the plainest of languages.
ArXiv.org has a copy of the BART paper:
http://arxiv.org/pdf/0806.3286
They are pretty good marketers actually; I have been to one of their talks, hosted by Andrew. I pointed a friend to this method; apparently, it's missing a predict method.
An application paper in a marketing journal is fine, and Arxiv is fine also, but I'm surprised not to see a paper describing the key ides of the method in a statistics journal.
It appears that the package is not only lacking a predict method, it also doesn't handle the normal formula + data.frame interface. The input and output are pretty simple so it shouldn't take a whole lot of wrapping to remedy that.
Anybody who really wants to try it out should be able to get on with it pretty easily.