Hello,
I am working on an influence diagram to identify the optimal policy on multiple decision nodes.
Can you provide further information about the computational time of the algorithm by Cooper (1988) [section 5.7.7.1 of GeNIe manual]?
Specifically, are you aware of any research investigating the computational time of the algorithm based on the number of decision nodes?
Alessandro
Influence diagram policy evaluation
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Influence diagram policy evaluation
In both influence diagrams and Bayesian networks, the number of nodes is not sufficient to estimate the computational time. The complexity depends on number of nodes, number of outcomes and the topology of the network.Specifically, are you aware of any research investigating the computational time of the algorithm based on the number of decision nodes?
Re: Influence diagram policy evaluation
Thank you, are you aware of any study investigating the computational time according to these factors?
-
- Site Admin
- Posts: 430
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Influence diagram policy evaluation
If you are interested in the complexity of the Cooper's algorithm, you should look at the complexity of probabilistic inference and the factors that influence it. There is a sea of literature on this topic. Cooper's algorithm, as implemented in GeNIe and SMILE, first converts the ID to a Bayesian network and then performs as many updates as are needed to evaluate all possible policies. The number of possible policies is roughly a product of the number of decision options in each of the decision nodes and each chance node that is observed before a decision node. I hope this helps.
Cheers,
Marek
Cheers,
Marek
Re: Influence diagram policy evaluation
Perfect, thank you!
Alessandro
Alessandro