Incorrect Diagnostic Value calculation ?

[Mohsin Beg]

Hi,

I think I may be hitting a SMILE/Genie issue.

In the attached zip (opening attached model and the case file)
when I set
"t24:found" as the pursued fault (with current posterior of 0.89),
"o168: shows a diagnostic value of 0.01.

Now when I set "o168" = notFound, the posterior
of "t24:found" changes to a posterior of 0.68.

I would tend to think that if the diag value is so low,
how can an additional evidence change the posterior by so
much. I think this may be a bug, no ?

Sincerely,

-Mohsin

[marek [BayesFusion]]

I think I may be hitting a SMILE/Genie issue.

In the attached zip (opening attached model and the case file)
when I set
"t24:found" as the pursued fault (with current posterior of 0.89),
"o168: shows a diagnostic value of 0.01.

Now when I set "o168" = notFound, the posterior
of "t24:found" changes to a posterior of 0.68.

I would tend to think that if the diag value is so low,
how can an additional evidence change the posterior by so
much. I think this may be a bug, no ?

Mohsin,

I wasn't able to reproduce your problem exactly, as you must have had some evidence observed before you performed the sequence above. I'm getting different posteriors and VOI for no other evidence observed.

In general, however, VOI computes *expected* change in the posterior. This means that the possible change for each case (i.e., possible observed state of the observation) is weighted by the probability of that observation. It is possible in individual cases that the observation state that changes the posterior a lot is very unlikely and, hence, brings down the computed VOI value. Could this be happening in your case? I am not excluding a bug in the software -- while GeNIe and SMILE are quite reliable, there is no software that is free of bugs.
Cheers,

Marek

[Mohsin Beg]

Hi Marek,

Unfortunately, I am still able to reproduce the problem at will. Please see attached screenshots (in zip).

And my genie (running on MS Vista) version is "2.0.3092.0 built on 6/19/2008"

Sincerely,

-Mohsin

[marek [BayesFusion]]

Unfortunately, I am still able to reproduce the problem at will. Please see attached screenshots (in zip).

And my genie (running on MS Vista) version is "2.0.3092.0 built on 6/19/2008"

This is not what I meant -- I was really asking for a sequence of observations that led to the situation. Unless, of course, it happens lal the time, under every circumstance. Your screen shots are all identical, by the way. Cheers,

Marek

[marek [BayesFusion]]

OK, now I undestand -- you had the case saved in the attached library. I am unable to check the a-priori probability of observing o168. I suspect that "not found", the outcome that changes the probability of t24 is unlikely. Also, the change that happens in the probability is smaller than it feels. Please note that it is ony 30% of the probability. Not much. What I want to say is that the results do not worry me too much. There is a possibility of a bug but I still find it unlikely . I hope this helps.
Cheers,

Marek