Re: Global consistency problem.

From: Warwick Harvey <wh_at_icparc.ic.ac.uk>
Date: Tue 05 Aug 2003 06:47:50 PM GMT
Message-ID: <20030805194749.X18432@tempest.icparc.ic.ac.uk>
On Tue, Aug 05, 2003 at 01:04:42PM -0400, Xiaohua Kong wrote:
> Hi Joachim,
> 
> Thanks for you answer.
> I am a new user in CLP. My target is to decide the range of a variable (say
> Va in my small example), so that there exists solution for the CSP problem.
> That's why the solution with the range of the variable is better than
> discrete values (such as, by using labeling). Is there any way I can do to
> find the range of a variable if the constraints are linear?

Hi Xiaohua,

If your constraints are linear, most likely you will be better off using the
eplex solver (see the constraint library manual) rather than the ic one.
The eplex solver interfaces to an external simplex-based linear constraint
solver (XPRESS-MP is available with ECLiPSe, but you can also use CPLEX if
you already have it).  Such a solver can fairly efficiently find an optimal
solution to any linear objective function you provide, so you can use it to
find the minimum and maximum values of a variable.  See in particular the
section "Probing Using a Different Objective", which talks about the
eplex_probe/2 predicate.

If you really want to use the ic solver for some reason (e.g. you have some
nonlinear constraints), then you can use the general branch_and_bound
library (see the reference manual).  This can also be used to optimise a
function, but is likely to be much less efficient (its advantage is that it
can optimise any function for any solver, given a suitable labelling
procedure).

Cheers,
Warwick
Received on Tue Aug 05 19:48:18 2003

This archive was generated by hypermail 2.1.8 : Wed 16 Nov 2005 06:07:25 PM GMT GMT