Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.
Sign up to join this community
Assume that I have a belief network with a set of nodes.
In order to create a valid junction tree I have to moralize the graph. Assume now that I have nodes with more than 2 parents (e.g 3 parents) then I have to link each of the parents. After doing that, the graphical models contain some unnecessary cliques. Can I use only a part of the cliques for the Junction Tree Creation? Or I must use every clique in the junction tree (e.g., a case with 8 nodes and 8 cliques)?
You only need to use maximal cliques. That is, if you have two cliques {a,b,c} and {a,b}, you only use {a,b,c}. Other than that, yes, you might generate unnecessary cliques, but how many depends also on the triangulation strategy (minimum fill-in is an NP-hard problem, so you won't really be able to do it perfectly in all cases).
I'm trying to implement the junction tree algorithm in Ruby.
I found that after moralizing, triangulating, and creating the final cliques by variable elimination, the final cliques contained all of the cliques from triangulated graph, but the maximally connected cliques in the triangulated graph were combined into one clique in the junction tree.
I'm new at this, but I don't think you can drop cliques along the way.
