2011/10/19 Geoffrey De Smet <ge0ffrey.spam@gmail.com>


Op 19-10-11 15:00, Guilherme Kunigami schreef:


In this use case, that is probably a bad idea in my experience. Why? Well I hope this makes any sense:
You need to allow the optimization algorithms to break it now and then to tunnel through a bad search space into another good search space.
If it doesn't, don't worry.

Hmm, I think I understood it. Allowing infeasible solutions may help to scape from local minima in the space of feasible solutions for example.
Yep :)
rule "Avoid conflicting activities"
when 
Assignment($room1 : room, $act1: activity, $id : activity.id)
Assignment(room== $room1, room != null, $act2 : activity, activity.id > $id)
Conflict(act1 == $act1, act2 == $act2)
I would put Conflict first. But try it this way too and let me know which works better ;) I don't know.
Stated differently: Instead of checking every 2 simultaneous assignments if they are a conflict,
I would check if every 2 conflict assignments are simultaneous (like in examinationScoreRules.drl).


Ok! I will perform some stress tests to verify which one works better.
Nice, please report your results to this mailing list. It doesn't matter if they are worse, better or equal: it's interesting to know.

Look for "stepLimit" in the examples to see how I do very short stress tests when adding extra constraints.


I've made a test with each model limited to 70 steps. I've attached a graph comparing both runs using drools planner benchmark. 
It  seems that using Conflict first is indeed faster :)

 
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users