Question about influence arcs entering a decision node

The front end.
Post Reply
CSPrice
Posts: 2
Joined: Mon Jul 09, 2018 11:26 am

Question about influence arcs entering a decision node

Post by CSPrice » Mon Jul 09, 2018 11:34 am

Hi there,

I am using an academic version of GeNIe (ver 2.1). I want to create a decision network where chance nodes influence the decision taken. However, when I add an arc from a chance node to a decision node, it treats the arc as an information arc and not an influence arc.

I notice that the help feature says: “Arcs coming into decision nodes have a different meaning. Because decision nodes are under decision maker's control, these arcs do not denote influences but rather temporal precedence (in the sense of flow of information). The outcomes of all nodes at the tail of informational arcs will be known before the decision will need to be made. In particular, if there are multiple decision nodes, they need to be all connected by informational arcs. This reflects the fact that the decisions are made in a sequence and the outcome of each decision is known before the next decision is made. Informational arcs are drawn as dashed lines.”

I have been using an older version of Hugin, and that software allows me to link chance nodes to decision nodes as influence arcs. When I imported my network into GeNIe, it no longer supported this.

Is there a way of making the chance nodes influence the decision in GeNIe? How do I model “Factors A, B and C influence the decision” in GeNIe?

Thanks, & kind regards
Sue

marek [BayesFusion]
Site Admin
Posts: 271
Joined: Tue Dec 11, 2007 4:24 pm

Re: Question about influence arcs entering a decision node

Post by marek [BayesFusion] » Mon Jul 09, 2018 12:58 pm

Hi Sue,

I have looked at your query but I am somewhat confused about a possibly different meaning of arcs coming into decision nodes in Hugin. I don't know Hugin well but the formalism of influence diagrams (and this should be shared between Hugin and GeNIe) assigns the meaning to decision nodes as nodes/variables that are under complete control of the decision maker. Nothing but the decision maker can determine their states. The decision maker picks the states of the decision nodes based on the expected utility computed by the diagram. Arcs entering decision nodes are information arcs and they essentially denote that the states of the variables coming in will be known by the time the decision will be made. I am standing behind GeNIe's meaning and implementation as correct and classical in terms of their meaning with my reputation as a decision analysis theorist and practitioner.

Hugin's implementation of influence diagrams must be correct as well but they quite possibly have introduced shortcuts that allow you to specify a problem in a slightly non-standard way. If you tell me what you want to accomplish, I will be happy to tell you how to do it. Your statement that there are variables that influence a decision does not sound theoretically correct, as decisions are under complete control of the decision maker. The decision maker will pick (or at least should pick, if he/she is rational) a decision that gives the highest expected utility and indirectly other variables will play a role here but they are not influencing the decisions directly. Please let me know what you want to accomplish (perhaps a concrete example will help) and I will be glad to show you how to model it.
Cheers,

Marek

CSPrice
Posts: 2
Joined: Mon Jul 09, 2018 11:26 am

Re: Question about influence arcs entering a decision node

Post by CSPrice » Tue Jul 17, 2018 10:09 am

Hi Marek,

Thanks for your response.

An example would be: if one is trying to decide whether one needs to take protection against rain (e.g. umbrella, raincoat, both or none), if you believe there will be rain but no wind, you'd take an umbrella only. For rain with wind, it would be raincoat or both. etc. The decider's belief of what the weather may be like may be wrong (hence my thinking that there should be a chance node or two entering the decision node).

For such a scenario, how would you model this?

Thanks, and kind regards
Sue

marek [BayesFusion]
Site Admin
Posts: 271
Joined: Tue Dec 11, 2007 4:24 pm

Re: Question about influence arcs entering a decision node

Post by marek [BayesFusion] » Tue Jul 17, 2018 10:36 am

Hi Sue,

I am enclosing two models for your problem. Model1.xdsl models your problem. I have invented all the numbers (probabilities and utilities).
Model2.xdsl is just to show you the difference between not knowing for sure what will happen (Model1) and knowing the states of Rain and Wind (Model2). I hope this will help. Please let me know if you have any questions.
Cheers,

Marek
Attachments
Model2.xdsl
(2.59 KiB) Downloaded 112 times
Model1.xdsl
(2.67 KiB) Downloaded 114 times

Post Reply