Dynamic Bayesian Networks

[Fernando Pasquini Santos]

Hi,

I looked for some tools for training Dynamic Bayesian Networks and I saw that pebl doesn't handle DBN training. I saw also that some other tools are just used to make inference about DBNs, but I didn't find anything that trains DBNs, and is implemented in Python.

So I was thinking if I just needed some short modifications in some parts of pebl to handle this. What if I just modify log-likelihood calculations, in scores, to look for samples in different time-slices? I think this won't be too difficult.

Of course that DBNs permit graphs to have cycles (as an edge to a variable to another indicates a relationship from 't' to 't+1'); and some other adaptations.

My question is: is someone already working with this? If not, I'm willing to make this contribution to the code.

Fernando