I need to be able to create non-linear, non-gaussian CPDs (let's call them Functional CPDs) for continuous nodes in BBNs and DBNs. I will also need to implement inference functionality on these nodes using any of the particle filters. Both of these tasks seem doable, and I have already started working on some of the code.
It would be nice if the learning code could handle functional nodes as well, though I don't understand enough about the learning process to say whether or not that will be possible.
Log in to post a comment.