Re: [eclipse-clp-users] fd_sets sameset negation

Thanks Matthew for your hint.

The solver really obviously doen's support negation like this. I helped
myself out by describing what I want using intersection and cardinality
constraints:
If two sets with cardinality 2 shall not be equal, then simply the
intersection must have a cardinality of lower or equal than 1. So I can
write:
            #(Pair1 /\ Pair2, I),
            I #=< 1

This should do the trick I guess. Note, this only works for my case because
I know that the sets have a cardinality of exact 2.


On 11 January 2011 15:34, Matthew Skala <mskala_at_cs.toronto.edu> wrote:

> On Tue, 11 Jan 2011, Oliver Shycle wrote:
> > "sameset(Set1,Set2)" that makes the two sets the same. But how does the
> > syntax look like, when I want the two sets to be different? If I use "neg
>
> The solver probably doesn't support such constraints.  Note that solving
> the negated version of a constraint may in general be much harder than
> solving the non-negated version; it's not a symmetrical situation by any
> means, so you can't generally expect that a given solver will support the
> negated version of each constraint it does support.  You should also be
> aware that problems that use these constraints may take much longer to
> solve than problems that don't.
>
> When I ran into a similar situation, where I wanted a "neither of these
> sets is a subset of the other" constraint in ic_sets, I ended up
> adding a customized constraint like this:
>
> set_nsub_cst(P,Q):-var(P),!,suspend(set_nsub_cst(P,Q),2,P->inst).
> set_nsub_cst(P,Q):-var(Q),!,suspend(set_nsub_cst(P,Q),2,Q->inst).
> set_nsub_cst(P,Q):-lists:subset(P,Q),!,fail.
> set_nsub_cst(P,Q):-lists:subset(Q,P),!,fail.
> set_nsub_cst(_,_).
>
> It shouldn't be difficult to create something similar for "set not equal,"
> depending on how much propagation you need.  The above code doesn't really
> do any propagation; it waits until the sets are fully instantiated before
> testing them.  On the other hand, with this kind of constraint there may
> not be much propagation possible anyway, and it worked well in practice
> in my application.
> --
> Matthew Skala
> Postdoctoral Fellow, University of Manitoba
> mskala_at_cs.umanitoba.ca
>