[gecode-users] variable selection heuristics

Guido Tack tack at ps.uni-sb.de
Fri Nov 6 15:48:25 CET 2009


Dear Alejandro,

there's no direct support for restart-based search, but it's easy to  
do with the search engines we provide.  Just use a FailStop object  
with a DFS engine, and when next() returns NULL, check whether the  
engine was stopped or search is exhausted.  If it was stopped, reset  
the engine and restart, setting a new limit on your FailStop object.

We've been looking into IBS as well, but, at least currently, I can't  
see a way to implement it in a non-intrusive way.  It's slightly  
difficult to do IBS in combination with recomputation - you have to  
measure the impact of a single choice, but with recomputation, you  
most of the time compute a fixpoint of more than one choice.  Also,  
keeping track of domain sizes is more intrusive than a simple failure  
count.  Anyway, we'll think some more about it, maybe we can come up  
with a solution that is general and elegant enough to be included in  
Gecode.

Cheers,
	Guido

Alejandro Arbelaez wrote:

> Dear Gecode developers,
>
> I've seen that the last gecode version includes variables' weighted  
> degree information which i think is a great new feature of gecode,  
> because now it will be possible to use the current state-of-the-art  
> variable selection heuristics (dom/wdeg) in gecode. I don't have all  
> the details about this new release, but I'd like know if this  
> release also includes a restart-base search engine. (i.e, restarting  
> the search engine when some cutoff limit in the number of failures  
> is met). I know this is not difficult to implement but would like to  
> know if this is already available.
>
> And my last question, are you considering to include Impacts [1]  
> into gecode (I mean in the near future)?
>
>
> [1] P. Refalo. Impact-based search strategies for Constraint  
> Programming, In CP 2004.
>
>
> Cheers,
> Alejandro
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users





More information about the gecode-users mailing list