Hi ,
urgent question: If I use conditional costs for a test in the diagnosis tool that means that the cost of my test is depending on the outcome of another nodes.
I expected that the expected cost will be calculated and weight against the information gain depending on the probabilities of these nodes.
So I expected that the conditional cost will influence the diagnosis result.
But even if I change the slider for the cost/entropy ratio the diagnostic values dont change.
It seems that the conditional cost only influence the diagnostic result if I set evidence for the respective nodes. But also without setting evidence there should be an effect if the outcome probabilities change, but there isnt. Why !!??
How can I model my net that these conditional cost are taken into account?
best regards
samoth
Conditional Cost in Diagnosis Tool
-
- Site Admin
- Posts: 1422
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Conditional Cost in Diagnosis Tool
Actually, it's even worse. If you run diagnosis with conditional costs and the cost parents have no evidence set the program behavior is undefined (the debug build asserts). If you want to continue with test costs, I recommend setting the 'mandatory' attribute in the properties of cost parents to true.samoht wrote:It seems that the conditional cost only influence the diagnostic result if I set evidence for the respective nodes. But also without setting evidence there should be an effect if the outcome probabilities change, but there isnt. Why !!??
This is definitely a bug and will be fixed in future releases. Can't give you any firm estimates though.