forum.hugin.com

User Forums => Research & Development => Topic started by: pibar on September 01, 2011, 16:07:45

Title: Learning algorithm
Post by: pibar on September 01, 2011, 16:07:45
Hola,

I just discover the new learning algorithm "Greedy-search and score". It provides better results in my application. My question is: What is the reference of this algorithm? How can I read about it? who is the author?

I notice that chapter 11 of the reference manual is not actualized with this new algorithms.

Congratulations for these new learning algorithms

Pablo I.
Title: Re: Learning algorithm
Post by: Martin on September 05, 2011, 15:58:07
The search algorithm repeatedly refines the model. At each step the algorithm greedily chooses to add, remove or reverse an edge, selecting the operation that maximizes the overall score.
Each step yields a refined model with a better BIC score than the previous. The algorithm terminates when no operation improves the overall score.

You can find an explanation of the BIC and log-likelihood scores in any bayesian network textbook. (e.g. http://www.hugin.com/technology/publications/bnid/)