Accessing cluster joint probabilities after inference

[f_allen]

I'm interested in running a modified version of EM to learn the parameters of a series of separate (different) bayesian networks with common parameters. It doesn't look like I can use the existing EM functions for this.

So in order to do this, I think I need to access the values of P(X,U|evidence) where X is a node variable and U is its parent variables, for all nodes as you would in standard EM on BNs. As I understand it, doing a clustering algorithm (e.g. DSL_ALG_BN_LAURITZEN) would result in values for all these joint probabilities being in the final clusters before the marginals are computed, but the NodeValue after running the algorithm only provides the final marginals. Is it possible to somehow access the joint probabilities from the clusters after the algorithm has run? I'm guessing the in-built em algorithm must do this somehow but I've not yet come across a way to do it.

Thanks for your help!

[shooltz[BayesFusion]]

I'm guessing the in-built em algorithm must do this somehow but I've not yet come across a way to do it.

EM does have the internal access to the SMILE's clustering implementation. We don't expose this functionality in publicly available library, as it's considered to be subject of change.