Good Morning All,
I was wondering if it's possible to get the combination of variables/factor/attributes that gives the highest Probability in the network for a specific target variable ?
Can we get the best combination of variables/ factor/ attributes that maximises/ minimize the expected utility ( node value).
Cheers
Didier
combination of variables for Highest Probability
-
- Site Admin
- Posts: 430
- Joined: Tue Dec 11, 2007 4:24 pm
Re: combination of variables for Highest Probability
Didier,
You may be referring to what is known as the MAP problem (Maximum A-posteriori Assignment) in Bayesian networks. GeNIe has an algorithm implementing this implemented (Annealed MAP in the Network menu).
In the second case, GeNIe calculates the combination of values of decision variables that maximize the expected utility. Is this what you want?
Cheers,
Marek
You may be referring to what is known as the MAP problem (Maximum A-posteriori Assignment) in Bayesian networks. GeNIe has an algorithm implementing this implemented (Annealed MAP in the Network menu).
In the second case, GeNIe calculates the combination of values of decision variables that maximize the expected utility. Is this what you want?
Cheers,
Marek
Re: combination of variables for Highest Probability
Marek,
does the optimization considers the evidence as well ?
Thanks again.
Didier
It was about the MAP problem indeed : Excellent, this is good new !!! does the the MAP consider the evidence as well ?you wrote " You may be referring to what is known as the MAP problem (Maximum A-posteriori Assignment) in Bayesian networks. GeNIe has an algorithm implementing this implemented (Annealed MAP in the Network menu)"
This is what I would like to have indeed !!!! How we can get the optimun combination variables from the Genie interface ?you wrote In the second case, GeNIe calculates the combination of values of decision variables that maximize the expected utility. Is this what you want?
does the optimization considers the evidence as well ?
Thanks again.
Didier
-
- Site Admin
- Posts: 430
- Joined: Tue Dec 11, 2007 4:24 pm
Re: combination of variables for Highest Probability
Didier,
MAP considers the current state of evidence.
The second question pertains to influence diagrams. When you update a diagram, you can examine the results in the decision nodes (or the utility nodes). The optimal decision is the one that has the expected utility in bold. It is the highest value of the utility. Situation gets more complicated when you have multiple decisions -- in that case you see the next optimal decision and then you can also see what the next optimal decision is conditional on what action you have taken in the previous step. In any case, it is always the decision for which the expected utility is in bold.
Cheers,
Marek
MAP considers the current state of evidence.
The second question pertains to influence diagrams. When you update a diagram, you can examine the results in the decision nodes (or the utility nodes). The optimal decision is the one that has the expected utility in bold. It is the highest value of the utility. Situation gets more complicated when you have multiple decisions -- in that case you see the next optimal decision and then you can also see what the next optimal decision is conditional on what action you have taken in the previous step. In any case, it is always the decision for which the expected utility is in bold.
Cheers,
Marek
Re: combination of variables for Highest Probability
Marek,
Thanks for your message.
I made some experimentation following your message. It looks working fine.
Cheers.
Didier
Thanks for your message.
I made some experimentation following your message. It looks working fine.
Cheers.
Didier