[gecode-users] Two questions about finite set variables

[Guido Tack]

Patrik Haslum wote:
> Q1: Propagation for the weights constraint. I've posted the  
> constraint:
[...]
> So, it seems that the upper bound on the cardinality of deleted has  
> been
> propagated to total_cost (it is at most the sum of the 3 largest  
> costs),
> but the lower bound has not (total_cost can not be smaller than the  
> sum
> of the 2 smallest costs, which is 2, yet the lower bound remains  
> zero).
> This is the case also if I change the upper bound on cardinality (for
> example, if I don't put one at all, the upper bound on total_cost
> becomes 60, the sum of all the costs, but the lower bound stays zero).
>
> Is there an explanation for this behaviour?

Yes, that kind of inference was simply missing in the propagator.  The  
trunk version should do it now.

> And, more importantly, is
> there any way to change it? Since I'm trying to solve an optimisation
> problem, it is of course the lower bound on cost that is important.

You could move to the trunk version of Gecode, or back-port the  
weights propagator (which shouldn't be too hard, you just have to take  
everything from the propagate method).

> Q2: When branching on in/not-in for a potential element of SetVar,  
> what
> is the order in which the alternatives are tried? (i.e., does it  
> explore
> the in or out branch first). And how do I change it to do the  
> opposite?

The left (first) branch includes the element, the right branch  
excludes it.  In the Gecode trunk, you also have the opposite option.

Cheers,
	Guido