Sorry i wasnt more clear. There is two aspects to this. The constraint
is pretty expensive to propagate ( this is a different constraint than
the one i send you the paper about btw, im using Gecode for two
separate projects ), for this i dont care if its placed at the very
end. That property is just something i wanted in order to tell whether
the constraint im adding is pruning values that would otherwise not be
pruned in that search node. Ie, if the other constraints have
stabilized and i then prune some values, i know these values would not
have been pruned by the other propagators unless a branching occurred.

The other constraints involved are distinct's and they hopefully dont
provide a cost of PC_CRAZY_HI :)

Regardless, this concept of staged propagation sounds interesting, i
will take a more thorough look at the paper.

Regards,
Peter Tiedemann

On 10/04/07, Christian Schulte <[EMAIL PROTECTED]> wrote:
Well, you cannot rely on that no other propagator does it. At least not in
real life...

What are you trying to do? If you have a propagator composed from a cheap
and an expensive stage you might want to consider staged propagation. Just
see the paper I mentioned before. We use it regularly in Gecode to great
benefit.

Cheers
Christian

--
Christian Schulte, http://www.imit.kth.se/~schulte/

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf
Of Peter Tiedemann
Sent: Tuesday, April 10, 2007 3:40 PM
To: [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: Re: [gecode-users] Ordering of propagators


> What do you mean with how the cost function is used? Could you
> elaborate, please?

I was just wondering how the cost affected the ordering, but it seems that
this is explained in the reference you just gave me. From a quick scan of
it, it would seem that if i choose the highest cost (and no other
propagators do the same), then i am in fact 100% sure that my propagator
will only be asked to propagate when all other propagators have stabilized,
which is exactly the effect i was hoping for.

Thanks for the quick answer, you guys really have short response times :)

Regards,
Peter Tiedemann



_______________________________________________
Gecode users mailing list
[EMAIL PROTECTED]
https://www.gecode.org/mailman/listinfo/gecode-users

Reply via email to