[Coin-discuss] Branch and price restore feasibility dual rays
Matthew Galati
magh at lehigh.edu
Sat Jul 24 08:29:27 EDT 2004
Hi Jakob,
I brought up the same issue back in December.
http://www-124.ibm.com/pipermail/coin-discuss/2003-December/000730.html
I don't use CPX much, so I never bothered investigating this more. But,
if I recall, I think the OsiCpx getDualRays is wrong.
Try switching the OsiCpxSolverInterface::getDualRays to use CPXs direct
call:
CPXdualfarkas(getEnvironmentPtr(), getLpPtr(), ray, NULL);
I think you also need to negate it, as this is what BCP expects:
transform(ray, ray + m, ray, negate<double>())
Let me know if this works and if not, send a small isolated example.
COIN folks, can we switch to a version of mailman that provides
searchable archives?
Matthew
>hi,
>
>I am looking again at my problem from february, where I cannot use the
>dual ray from OsiCpxSolverInterface to generate a variable breaking
>infeasiblity, whereas the ray from OsiClpSolverInterface works well.
>Until now I used Clp which worked well,
>but I would like to compare to cplex.
>
>I am applying B&P to a 2D bin packing problem, each variable is
>a pattern, and each row corresponds to an item (it guarantees that the
>item is used).
>OsiClp returns a dual ray having one entry set to a high value, and the
>others to zero, it therefore gives me the index of a missing item of
>the solution (i.e. adding a pattern containing this item can restore
>the feasibility).
>OsiCpx returns a dual ray having several entries set to 1, or 0.5, but
>the entries correspond to items which are allready available in the
>solution, therefore I cannot generate a variable breaking infeasibility
>using this information.
>
>thank you for your answers (also for the answers from february)!
>
>best wishes
>jakob puchinger
>_______________________________________________
>Coin-discuss mailing list
>Coin-discuss at www-124.ibm.com
>http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss
>
>
--
Matthew Galati -- Optimization Developer
SAS Institute -- Analytical Solutions / Operations R & D
Office R4112 O: 919-531-0332 F: 919-677-4444
URL: http://sagan.ie.lehigh.edu/mgalati/
More information about the Coin-discuss
mailing list