Probabilities for several nodes

[Suchard]

Goodday,

I've got a question concerning probability values: my BN consists of many nodes each of them with two states ("OK" / "not OK") that are interlinked in different manners.

After having filled in the CPTs, I can, of course, easily find out what the occurance probabilities for the states of one node at a time are. But how can I find out the probability that two (or more) specific nodes are in the state "not OK" at the same time?

As known from fault tree analysis, two probabilities can only be multiplied if they are independent which, however, is not the case for my concerning nodes.

I would be very happy if someone could quickly give me a hint as I'm pretty close to deadline..


Thank you in advance,
Suchard

[Martijn]

Hi,

You would need the joint probability distribution of the nodes you are interested in.
Unfortunately, it is not possible to acquire such a distribution in GeNIe directly, but it is possible to manually try and acquire what you need.
Don't know how feasible it will be to do it by hand though.

We've had this question before, for SMILE, our inference library, which is the heart of GeNIe.
Please see here: http://genie.sis.pitt.edu/forum/viewtop ... tion#p2989 for more information.

Best,

Martijn

[Suchard]

Thank you very much for your helpful answer.

Concerning the link you posted, I tried option number (2): "Add a "join node" to the network, with all the disease nodes as parents and one state for each combination and a deterministic definition."

If I understood it right, this is a very easy option for my problem as for the combination of two nodes (each of them with two states), the states in the deterministic join node are "OK / OK", "OK / not_OK", "not_OK / OK" and "not_OK / not_OK". After having filled in the CPT of the join node, I see the result right away.

Am I right?

Cheers,
Suchard

[Martijn]

hi Suchard,

Yes, that's how it works.

Best,

Martijn