[Scip] Branch and Bound Bi-objective
Gregor Hendel
hendel at zib.de
Mon Mar 16 14:34:40 CET 2015
Dear Emmanuel,
sorry for the late reply. The use of customized relaxations requires the
implementation of a relaxation handler. A quick starting guide on how to
do this can be found in our HowTo section on relaxation handlers
<http://scip.zib.de/doc/html/RELAX.php>.
The crux is, of course, how to handle additional objective coefficients
for every variable. This can be currently managed, eg, by entering
diving mode of SCIP, changing all variable objective coefficients, and
resolving the diving LP.
Your relaxation should then also have a mechanism to store every node's
lower bound according to the
additional objective.
As far as I know, multiobjective problems are usually solved as a series
of weighted combinations of the objective functions. The generation of
such a sequence is easy to achieve by changing the objective coefficients
for all variables, and resolving the problem with SCIP. (Since primal
solutions stay feasible, you might consider transferring them from one
solving round to the next, as well).
If everything works as planned, the next SCIP release will contain more
refined reoptimization techniques and a multiobjective solver as an example.
Hope this helps,
Gregor
Am 11.03.2015 um 15:54 schrieb Emmanuel Kieffer:
> Hello,
>
> I would like to know if it is technically possible to use SCIP to
> perform a Bi-objective Branch and Bound?
> If yes, I need to solve several relaxed problem at each nodes. So far I
> know, it's possible to deactivate the standard LP relaxation and use a
> custom one but how is it work?
>
>
> Thank you
>
> Best regards,
>
> Emmanuel
>
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20150316/0fe3da68/attachment.html>
More information about the Scip
mailing list