Influence Diagram

The front end.
Post Reply
parvaneh2
Posts: 2
Joined: Tue Mar 12, 2013 3:19 pm

Influence Diagram

Post by parvaneh2 »

Hi, I am currently working on the decision making of the ship crew in case of fire. I want to use ID for the decision making process. I have some nodes that are the factors that affect the decision node. How should I design my network? Should I consider these factors as "chance nodes"? Shall I draw the arc from those nodes to the decision node? Is it possible in ID? Thanks in advance.
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Influence Diagram

Post by marek [BayesFusion] »

Well, you are asking a very general question, one for which I should direct you to a good decision analysis textbook or take a course in decision analysis. There must be some in your library -- just look for the term "decision analysis".

Generally, the answer is yes, you should model the factors as variables. Most of the time they will be uncertain. You will also need a value node. Arcs coming onto decisions are information arcs and mean temporal precedence. GeNIe implements full influence diagrams, so anything that you will see in the texbook should work in GeNIe.
Good luck!

Marek
parvaneh2
Posts: 2
Joined: Tue Mar 12, 2013 3:19 pm

Re: Influence Diagram

Post by parvaneh2 »

thank you so much for the reply. as you recommended, I read some books and topics in the "decision analysis" concept. Still I might have a few questions, one of them is about the diagram attached to my comment. I wanted to know, why aren't nodes "Carcinogenic Potential" and "Exposure Rate" directly connected to the decision node and two other nodes called "test" and "survey" which has almost the same variables are connected to the decision node? Can we just omit the other two nodes and connect nodes "Carcinogenic Potential" and "Exposure Rate" to the decision node?
Does the connection of the nodes "test" and "survey" mean that they are variables effecting the decision process?
It would be great if you answer my questions one by one, not generally, since it is very important for me. Thanks.
Attachments
Figure 03.20.xdsl
(4.94 KiB) Downloaded 336 times
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Influence Diagram

Post by marek [BayesFusion] »

Dear parvaneh2,

The nodes Test and Survey are going to be observed before the decision (Usage Decision) is mad, the nodes Carcinogenic Potential and Exposure Rate will not be observed before making the decision. The creator of this model most likely wanted to express the fact that neither carcinogenic potential nor exposure rate can be observed with certainty but you can conduct a test and a survey that will reveal some information about them. This is the correct (and pretty much standard) way of doing this. If you think you can observe both carcinogenic potential and exposure rate with certainty, yes, you can omit the Test and Survey nodes and draw arcs from Carcinogenic Potential and Exposure Rate to the decision.

Nodes that have arcs from them into the decision nodes affect the decision in the sense that the decision maker will know their states (dashed arrows denote this) but the decision is more affected by maximization of the expected utility and the observations change the situation. Please note that both nodes indirectly affect Cancer Cost.

I hope this helps.
Cheers,

Marek
Post Reply