How to find joint marginal distribution?

[gund]

When I update beliefs in an influence diagram, the values in the chance nodes are conditional marginal probability distributions (conditioned on the informational nodes and/or decision nodes).

Can I obtain joint marginal distributions by calling some method?

Basically instead of P(node A = state0), I want P(node A = state0 AND node B = state1). Do I have to manually multiply out the conditional marginal probabilities myself? I wrote the code to do it and it seems correct, but rather use a built-in method.

What about for Bayesian networks with marginal distributions?

[marek [BayesFusion]]

When I update beliefs in an influence diagram, the values in the chance nodes are conditional marginal probability distributions (conditioned on the informational nodes and/or decision nodes).

Can I obtain joint marginal distributions by calling some method?

Basically instead of P(node A = state0), I want P(node A = state0 AND node B = state1). Do I have to manually multiply out the conditional marginal probabilities myself? I wrote the code to do it and it seems correct, but rather use a built-in method.

What about for Bayesian networks with marginal distributions?

I'm afraid neither GeNIe nor SMILE have this capability. You will have to compute joint marginal outside of SMILE. The same holds for both IDs and BNs. Good luck!

Marek

[marek [BayesFusion]]

When I update beliefs in an influence diagram, the values in the chance nodes are conditional marginal probability distributions (conditioned on the informational nodes and/or decision nodes).

Can I obtain joint marginal distributions by calling some method?

Basically instead of P(node A = state0), I want P(node A = state0 AND node B = state1). Do I have to manually multiply out the conditional marginal probabilities myself? I wrote the code to do it and it seems correct, but rather use a built-in method.

What about for Bayesian networks with marginal distributions?

I'm afraid neither GeNIe nor SMILE have this capability. You will have to compute joint marginal outside of SMILE. The same holds for both IDs and BNs. Good luck!

Marek

[gund]

Ah ok thanks. Not a big problem.

Edit: I noticed in GeNIe there's a button called P(e). Looks like for BNs, you can set the evidence as the joint marginal probability you are interested in, and it will calculated the P(evidence).

[marek [BayesFusion]]

Ah ok thanks. Not a big problem.

Edit: I noticed in GeNIe there's a button called P(e). Looks like for BNs, you can set the evidence as the joint marginal probability you are interested in, and it will calculated the P(evidence).

Yes, this is one way of computing individual elements of the joint p.d. Just watch out other evidence in the network, outside of the nodes for which you are computing the P(e) -- you will have to compensate for those (using essentially Bayes theorem).
Cheers,

Marek