Find best policy algo vs policy evaluation algo

[gund]

I have a test ID that using "policy evaluation" algo, gives me some decision A (correct one) as the decision with the highest expected utility, but using "find best policy", gives me decision B (last decision always) as the optimal policy.

Not sure why this would be the case.

Is there any condition such that this is correct behavior?

The test ID has 2 decision nodes.

My best guess of what happened is the 'find best policy' algo failed and gives the last decision as the optimal one by default.

[shooltz[BayesFusion]]

Can you post your influence diagram here?

[gund]

Here it is.

Sorry it's a little convoluted hehe. Messed around with it. I was trying to see if I could answer the question of VOI by phrasing it as another decision node.

Have another question.

How is value of information calculated? Does it use the policy evaluation algo or the find best policy algo. Because I think it could use either and I suppose the find best policy algo is faster?

[shooltz[BayesFusion]]

How is value of information calculated? Does it use the policy evaluation algo or the find best policy algo. Because I think it could use either and I suppose the find best policy algo is faster?

See this thread for more information on VOI:

http://genie.sis.pitt.edu/forum/viewtop ... hlight=voi

[shooltz[BayesFusion]]

I have a test ID that using "policy evaluation" algo, gives me some decision A (correct one) as the decision with the highest expected utility, but using "find best policy", gives me decision B (last decision always) as the optimal policy.
Not sure why this would be the case.

That's a bug in SMILE; it's likely that it's triggered by unusual structure of your influence diagram

We're investigating the issue.