Is there a way to incorporate constraints on other nodes when performing target optimization? Given that a genetic algorithm cannot directly handle constraints (they must be incorporated into a pseudo-objective function with penalty multipliers), this may impossible with the current configuration of Bayesialab, but is there any kind of "work around"? For example, maximize the target node subject to node A < X and node B > Y?
Quote 0 0
Dan
You're right, there is no way right now to introduce such constraints in the evaluation function of the genetic algorithm used for the optimization. The solution consists in using Constraint Nodes to express your constraints: http://library.bayesia.com/display/BlabC/Constraint+Nodes.In your example, you will have to use 2 constraint nodes, as illustrated below:
Quote 0 0
Apologies for resurrecting a dead thread here...I never really managed to get this to work, so I'm returning to it now.I want a constraint that is satisfied, i.e., "True", if the mean value of node A is less than or equal to a constant K. In the constraint node, as you suggested, I'm using the deterministic equation option. I've tried:?C1? = ?A? <= K?C1? = if(?A? <= K,True,False)and the same format but with B instead of A, where B is a function node set to MeanValue(A).None of these approaches seem work. Any idea where I'm going wrong?Thanks
Quote 0 0
Dan
Hi,I just designed the BBN below. It has continuous nodes and constraint nodes. C1 is defined as: ?MarketShare_1?>?MarketShare_4?C2 is defined as: ?MarketShare_2?>?MarketShare_11?C3 is defined as: ?MarketShare_3?>0.04Running Target Optimization on this network returns policies satisfying these 3 constraints.However, you cannot use a Function node as a parent of a Constraint node (we need to forbid it).
Quote 0 0
Thanks, I'll try that again to see where I made the mistake.
Quote 0 0