[gecode-users] How to stop a branching?

[Martin Mann]

mhhh.. what a pitty.. but yes, there might be problems with recomputation..

Some helper variable e.g. posted within the incremental posting might 
help but I doubt that anything gets faster by that.

Nevertheless, thanks a lot for the fast reply!

Martin



Christian Schulte schrieb:
> Right now, you can't stop branching. While just having an operation to do
> that is easy, but I do not see yet how one would get that to work with
> recomputation.
> 
> Did you see that you can actually post branching incrementally, maybe that
> could help. Check "Executing code between branching" in MwG. You could
> control whether a branching is posted or not by an additional variable. Just
> an idea, don't know whether that would help.
> 
> Christian
> 	
> 
> -----Original Message-----
> From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
> Of Martin Mann
> Sent: Friday, May 29, 2009 2:04 PM
> To: gecode user list
> Subject: [gecode-users] How to stop a branching?
> 
> Hi,
> 
> I would like to know, if it is possible to stop a branching (when 
> multiple ones are present). Thus, to ignore all remaining branches for 
> this specific one.
> 
> Maybe an example to make my question clear.
> 
> I have two sets of variables X_1, X_2 linked via several constraints. 
> Within my Space I post two brachning B_1(X_1) and B_2(X_2) such that 
> first all variables of X_1 are assigned before B_2 is considered.
> 
> My final goal is to enumerate solutions with different X_1 assignments!
> 
> My current solution (was implemented for Gecode 1.3.* and I am going to 
> reimplement in 3.*) is to formulate a COP instead of a CSP and to 
> implement the *constrain* function and to post a constraint that ensures 
> that the next solution is at least in one X_1 assignment different from 
> the current solution.
> 
> Thus, I get a new constraint per solution ... and I dont like that! ;)
> 
> But when looking at the problem, it would be sufficient to stop the 
> branching on X_2! I dont need to post new constraints but to cancel the 
> remaining branching of B_2 on X_2. This will lead to a backtracking in 
> B_1 and therefore ensures that the next solution differs in the X_1 
> assignment in at least one solution...
> 
> This seems to be much more efficient to me. Btw. I would end up with a 
> CSP instead of this strange COP hack.
> 
> So my question: Is it possible to stop the remaining branching on B_2 
> out of the (Gecode 3.*) box? Or do I have to implement/subclass my own 
> branching?
> 
> Thanks for your help,
> 
> Martin
> 
>