Suppose I have a network like the one shown in the new documentation for Target Optimization in version 6.0: http://library.bayesia.com/pages/viewpage.action?pageId=16319181How can I create a Pareto-frontier between total costs, and sales entropy, using target optimization? I want to be able to say, "For X amount of cost, I can achieve an uncertainty value of Y." I created a network with similar elements but can't produce good results:Profile Search Criterion - Function Value: Sales EntropyCriterion Optimization - Target Value: 100 (I normalized entropy so that the function node shows 100% if 0 entropy is attained; this is analogous to saying "minimize the uncertainty")Take Into Account the Resources - This is the part where I'm really stuck, as no matter what I use for min, max, or target resources, it seems to ignore those values.Weighting is set so that Target Value and Resources have the same weight (required for a basic Pareto-frontier)Any help would be greatly appreciated.
Quote 0 0
Dan
Unfortunately, there is no direct way to produce a Pareto Frontier with BayesiaLab. If I understand correctly, you want to use the best solutions that have been found during the optimization for building your Pareto Frontier. As per the Resources' settings, the Minimum and Maximum values are used to define the actual range of possible values your Resources can take (BayesiaLab can compute it directly with the Function Nodes). In the example shown in the Library, we get the min value by setting all the nodes to their first state (see below), and the max by setting them to their last state.
Quote 0 0
Hi Dan,Thanks for the reply.Yes, you are correct that I intended to create the pareto-frontier indirectly using target optimization. I had thought - erroneously it would seem - that I could use the "target resources" option to effectively tell the optimizer, for example, "For 100 resources, what is the best reduction in uncertainty I could achieve?" I could then repeat the optimization with target resources of, again, for example, 150, 200, 250, and 300, where I would expect to see a greater reduction in uncertainty with greater resources spent. These points I could then use to manually create a pareto frontier of uncertainty reduction vs cost.Is there no way to do this? I doesn't appear that using "min" and "max" resources can achieve this result. I tried to create something to similar effects with the decision and utility nodes, but the catch there is that "Gain" (uncertainty reduction) and "Cost" (resources), would have to have the same units, which is something I want to avoid.
Quote 0 0
Dan
Settings "target resources" should indeed help you achieving your goal. Perhaps the Genetic Algorithm does not succeed in finding a correct solution.Did you try increasing the resources' weight defined in the Weighting section?
Quote 0 0
Thanks, I'll play around with this some more.
Quote 0 0