Hi Michael.
I am a M.Sc. student on Bayesian networks at the University of Regina, Canada. As lots of students of BNs, I have tried to implement a simple exact inference library but got stuck because of the multiplication/marginalization of factors implementation.
I basically followed Koller & Firedman 2009's book (Algorithm 10.A.1) for multiplying and marginalizing factors, but the implementation (it doesn't matter the language) is too slow for even medium size networks (can take 15 min in a 50 nodes BN).
In the other hand, Jayes could answer a query in medium size BN in less than a second, which is very nice for experiment results. I believe this has to be with the implementation of multiplication/marginalization of factors.
I was wondering if you could give me more details by recommending a paper or giving some insights on how you have implemented these factor operations in Jayes, please. This would be helpful for further increments on Jayes as our research group uses the library.
Thank you.
Best regards.
