2008/4/22 Geoffrey De Smet <ge0ffrey.spam@gmail.com>:

Hi guys,

I finished 4th with my drools-solver implementation of the examination track:
http://www.cs.qub.ac.uk/itc2007/winner/finalorder.htm

There's much to learn from the other implementations for drools-solver.
For example, the winner is also using local search (as are most of the others), but he use meta-heuristics to run 3 phases instead of 1, selects his moves a lot smarter and just basically tweaked his configuration a lot better.

Surprisingly, I actually got the best overal result on dataset10: it will be interesting to find out why (could be tweaking luck though).
I 'll be identifying the features which drools-solver is missing and incorporating them soon.
Since the end of the competition, the trunk has already gained 10% performance (due to removing shadow facts etc).

I've also done an implementation of the curriculum course track after the competition ended, in less than 16 workings hours.
The mere development speed, maintainability and runtime performance (after initialization...) of DRL score rules is very very promising.

--
With kind regards,
Geoffrey De Smet

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