Does OpenBayes support creating/removing nodes to better model a partially observable structure? I better description of this might be http://www.cs.ubc.ca/~murphyk/Bayes/bnintro.html#learn under the "Unknown structure, partial observability" header.

Regards,
Chris

Chris,

The short answer is no. This is something i would really like to see added, and nothing in our BN structure precludes it from happening. But currently the package only has basic learning in it, though I would like to see this improved. What you are asking (if i am understanding murphy correctly) is really about learning the structure of a bayes net. Currently this kind of learning is completely unsupported (though it would be great if you wanted to investigate it).

The good news is that the current packages basic underlying representation of BN would allow you to easily add nodes to one part of a network without having to do any especially expensive network recreate operations.

We would love to have more people working on this package, let us know if you are interested.

I'd like to announce you that a student (François) is currently working on the SEM algorithm which allows to learn the structure and the parameters of a BN from observed data.

The final implementation will involve learning with 'unknown structure and partial observability'.

This new learning engine should be available by january more or less.

Hope this helps

Kosta

The EM and SEM algorithms are now available!

kosta