Given a densely connected Bayesian Network based on expert input, what is a good algorithm for looking for edges that could be removed? All the nodes are Gaussian.
I could discretize the variables and then use K2, but maybe there is another way?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
One way to do this, would be to remove the edges that are not in the minimal I-map of the network. When all the nodes are Gaussian, the minimal I-map can be determined by calculating the inverse of the covariance matrix. The zero elements in this matrix correspond to the absence of arcs.
An explanation can be found in the textbook "Probabilistic Graphical Models, principles and techniques" by Koller and Friedman. The original (Open Access!) paper is: Gaussian Markov Distributions over Finite Graphs; T. P. Speed and H. T. Kiiveri; Ann. Statist. Vol. 14, Number 1 (1986), 138-150. http://projecteuclid.org/euclid.aos/1176349846