Evaluating Decision networks (influence diagram)

[beambeam]

Hello all,

I am wondering if there is some algorithm to learn probability in Decision network or(Dynamic decision networks)
i mean is there any algorithm for generating the probabilities , the utilities...

in the examples listed in the literature, the values of probabilities and utility are setting by user
i want to implement an influence diagram and i want to know how generate the probabilities with automatic manner?

Thanks in advance.

[marek]

Good question. I wish I could give you an enthousiastic and positive answer, i.e., automcatic creation of decision models from data.

Here is what you can do now:

(1) Learn the uncertain nodes in your influence diagram from data using Bayesian network learning (there are several algorithms for that in GeNIe). An influence diagram is a Bayesian network enhanced with decision node(s) and utility node(s).

(2) Create your decision node(s) manually and connect them to the Bayesian network that you have learned. Performing this task automatically (learning what decision options you have from data) would be quite a challenge.

(3) Create your utility nodes and equip them with utility functions. There has been research on learning preferences but I believe it is still far from learning utility functions, so I believe there is no software that can do this.

I hope this helps.
Cheers,

Marek