Re: [rules-users] Proposal: Collect Case Studies

Hi folks, I used to buy some hardware from here and played with their configurator: http://www.alternate.de/html/includeStaticBig.html?treeName=KONFIGURATORE... (The page is in German but they have graphic icons around it's not hard to understand....) If you choose an amd based motherboard and an Intel CPU together, their web configuration will warn you and you have to re-choose. I do not really know how they implemented it but I thought a rule-based implementation is very suitable for such kind of thing. I ran into a combinatorial optimization problem recently. At the first thought, any Rete-based rule engine is not really the right tool for searching optimal solutions in a *huge* solution space. But when I started to implement my algorithm, it turns out that Drools can still possibly help me in two places. I decided to use Monte Carlo optimization algorithms, for my problem the most interesting candidates are evolution strategy and genetic algorithms. Before I start to randomly poll elements from the solution space, I could use drools to efficiently filter out the absolutely unwanted elements (this has its drawbacks too...depends on how tolerate to the end result the end users *want* to be). My filter condition changes very often (it would depends on the things like personal preference of a particular user, etc), so using drools for filtering makes sense. The filtering does effectively reduce the solution space. And then, drools could be used to calculate the global score after each poll from the solution space. If I use the genetic algorithm, the mixing rules could also be defined in a .drl file (which will be very easy to change, and yes, it needs frequent changes when I want to tune the system). The gene mixing could also be running in Drools' working memory. There might be some performance gain against straight iteration + data manipulation in a single java procedure. But I have not implemented it yet (will start coding from tomorrow), so far not sure about the performance gain. if anything turns out good, I'd be glad to report. Regards, Ellen N. Zhao On 7/18/07, Dr. Gernot Starke <gs(a)gernotstarke.de&gt; wrote: