Dynamic Bayes Nets

Dynamic Bayes nets are also known as DBNs or temporal Bayes nets.  They allow you to specify a Bayes net model which has "time delay" links, indicating that the value of the child node depends on the value of the parent at an earlier time.  Later you expand the DBN so that some nodes in the original net become several nodes in the expanded net, indicating the value of that variable at different points in time.  

DBNs can have directed cycles, as long as there is a delay link somewhere along each cycle.  Delay links can be used to model feedback.  Once it is expanded, it will no longer have cycles or delay links.

We improve Netica's DBN capability with each release, so if you are working with DBNs you should use at least version 5.02.  Download latest versions from our ftp site.

It is usually easiest to work through an example DBN first.  The Bayes net called "Bouncing", in Netica's “Examples” folder, is suitable for that purpose.

Steps for working with a Dynamic Bayes Net:

1. Create DBN

2. Generate time expansion

3. Compile and use

If you have any requests or suggestions for Netica's DBN feature, be sure to contact our support team, since we are actively improving this feature.

See also: The tutorial on DBN Bouncing Example