paper available: Efficient Approximations for the Marginal Likelihood...

[David Heckerman]

The following paper is available on the web at

http://www.research.microsoft.com/research/dtg/heckerma/heckerma.html

     Efficient Approximations for the Marginal Likelihood
        of Incomplete Data Given a Bayesian Network

              D. Chickering and D. Heckerman

                       MSR-TR-96-08

A Bayesian score often used in model selection is the marginal
likelihood of data (or "evidence") given a model.  We examine
asymptotic approximations for the marginal likelihood of incomplete
data given a Bayesian network.  We consider the well-known Laplace and
BIC/MDL approximations, as well as approximations proposed by Draper
(1993) and Cheeseman and Stutz (1995).  In experiments using synthetic
data generated from discrete naive-Bayes models having a hidden root
node, we find the Cheeseman-Stutz measure to be the best in that it is
as accurate as the Laplace approximation and as efficient as the
BIC/MDL approximation.


The paper also can be retrieved via anonymous ftp:

ftp-host: ftp.research.microsoft.com
ftp-file: pub/tech-reports/winter95-96/tr-96-08.ps

-David