You might be disabling "delta score calculation" with your score rules: https://github.com/triceo/DroolingTSP/blob/master/src/main/rules/tspScoreRules.drl Read http://downloads.jboss.com/drools/docs/5.1.1.34858.FINAL/drools-planner/html_single/index.html#d0e1530
Simulated annealing doesn't work well when it's difficult to find solutions which have no hard constraints broken. I am investing at a fix for that. Mix in a course-grained MoveFactory (but keep the current one too). Have you tried the benchmarker to try different tabu search configurations against each other? Property tabu 5 or 7 and Move tabu 7 or 11 usually work well. HTH Op 14-02-11 09:34, Lukáš Petrovický schreef: > Hello, > > I've been making myself familiar with Drools Planner and in the process > of doing so, I've hacked together a simple algorithm for optimizing the > Travelling Salesman Problem: > > https://github.com/triceo/DroolingTSP > > At this point, I get a 7 % improvement over a route with 71 stops, which > was previously assembled using a nearest-neighbor algorithm. The reason > why the improvement isn't bigger is that I'm stuck with Tabu search only > - for my use case, simulated annealing doesn't work at all. (Exits > forcefully after 1 to 7 steps, resulting in only a minimal improvement.) > > I'll be interested in any comments and suggestions you might be able to > provide. > > Best regards! > -- With kind regards, Geoffrey De Smet _______________________________________________ rules-users mailing list [email protected] https://lists.jboss.org/mailman/listinfo/rules-users
