[SCIP] Presolving after pricing variables
Gerald Gamrath
gamrath at zib.de
Thu Jul 25 21:29:46 CEST 2019
Hi Jan,
there are two options that you already mentioned which I would suggest
as well, but I did not try them myself:
1) copy the problem using SCIPcopy(), where you can set copypricers to
FALSE. Additionally, you need to loop over all constraints in the
subscip once and set them to be not modifiable via
SCIPsetConsModifiable(). SCIP can only perform very few presolving,
propagation, and separation steps if constraints are marked to be
modifiable, because in that case, SCIP needs to assume that new
variables will be added to the constraint later which might conflict
with the reductions performed now.
2) Again, set all constraint to be non-modifiable, disable the pricer,
and then perform a restart and solve the problem.
In both cases, you potentially need to re-enable features that you
disabled for the column generation phase.
I would probably go for the second option first, because it should be
easier to implement, but I am not sure if perhaps some flags can still
be set incorrectly in that case. If you still observe considerable
differences, please write again.
Best,
Gerald
Am 24.07.19 um 18:54 schrieb Jan Berling:
>
> Dear SCIP community,
>
> I have a binary optimization problem which uses column generation, i.
> a. because adding all variables is not possible due to memory
> limitations. However, solving takes a lot of time and so pricing is
> stoppedwhen a timelimit is exceeded, by setting the result pointer to
> SCIP_SUCCESS.
>
> SCIP continues solving the “final restricted master problem” but this
> takes a lot of time.
>
> I made an experiment giving the final transformed lp file to
> out-of-the-box-SCIP and the problem is solved more quickly, due to
> presolvers and cutting planes.
>
> Is it possible to restart the solving of the transformed problem after
> pricing is stopped and to enable presolving/cutting-planes (e.g. by
> copying the problem, enabling presolvers, restarting, setting SCIP
> stage...)?
>
>
> Kind regards,
>
> Jan
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20190725/3ec34154/attachment.html>
More information about the Scip
mailing list