sensitivity analysis

[mc]

If I built a network that is not an influence diagram can I make the sensitivity analysis with the addition of a Decision node "Sensitivity" as explained in the Tutorial 12? Thank you very much!

[marek [BayesFusion]]

If I built a network that is not an influence diagram can I make the sensitivity analysis with the addition of a Decision node "Sensitivity" as explained in the Tutorial 12? Thank you very much!

Yes, you can. Just add a decision node along the same lines as in the tutorial, update the network, and see the impact of this on the posterior probability distribution over the node(s) of interest. They should be indexed by the newly introduced decision node.
Cheers,

Marek

[mc]

Yes, you can. Just add a decision node...
Marek[/quote]

I tryed once and doesn't work very well. I had some problems with GeNIe and the software crashes. Moreover in the output window this message appear:
"Utilities not specified. Policies cannot be evaluated"
thus every time I work (e.g. introducing an evidence) and updating the network.
Therefore I tryed again adding an utility node (just to avoid problems...) and it works.
I am wondering if this change would affect the sensitivity analysis since now my network became a decision network .

Thanks.

mc

[marek [BayesFusion]]

I tryed once and doesn't work very well. I had some problems with GeNIe and the software crashes. Moreover in the output window this message appear:
"Utilities not specified. Policies cannot be evaluated"
thus every time I work (e.g. introducing an evidence) and updating the network.
Therefore I tryed again adding an utility node (just to avoid problems...) and it works.
I am wondering if this change would affect the sensitivity analysis since now my network became a decision network .

Wow! You managed to crash GeNIe. Would you be willing to send us a network that achived this result? I really mean it -- GeNIe should not crash no matter what.

Adding a dummy utility node indeed does the job -- the software does not complain now. It would be convenient to run the algorithm without a utility node. I have added this to our TO DO list.
Cheers,

Marek

[mc]

Wow! You managed to crash GeNIe. Would you be willing to send us a network that achived this result? I really mean it -- GeNIe should not crash no matter what.

Marek

Since I had again problem with my net to which I added decistion nodes to some variables to test the sensitivty here is my network, nothing special anyway..
I thought that the problem could be not having introduce the value node, what do you think?
mc