Building Probabilistic Graphical Models with Python

Steps in the LBP algorithm

We can summarize the steps in the LBP algorithm as follows:

• Assign factors to clusters
• Construct initial potentials (multiply all the factors in a cluster)
• Initialize beliefs at each cluster
• Repeat the message passing step: choose a cluster pair and pass a message between them
• Test whether the convergence has occurred or stopped after n repeats
• Summarize beliefs at each cluster by multiplying all the messages received with their initial beliefs as follows:
  

Here, the belief in the ith cluster is the product of initial belief and ...