[Scip] SCIP 3.1 is slower than 3.0?

[Gerald Gamrath]

Hi Alessia,

can you send me logs (including statistics) for the two runs?

One thing you can try is to set "lp/disablecutoff" to 0. Before SCIP
3.1, the cutoff bound was always applied in the LP solver, even if you
do branch-and-price and will use the dual solution to generate new
columns. It is still a correct algorithm if you use the dual values you
get when the objective limit was reached, but you will need to solve the
LP from scratch afterwards. Especially for large master LPs, this can be
quite expensive, so that we automatically disabled it now, but it is a
sort of stabilization and might pay off if your master LP is very easy
to solve and solving it from scratch doesn't hurt too much.

Best,
Gerald

On 12.03.2014 15:59, Alessia Violin wrote:
> Hello,
>
> I just passed to the new version of SCIP and tried to make a couple of
> tests, the new version seems to be significantly slower compared to
> the 3.0.1.
>
> I am using Cplex as the lp solver, and I had to compile with the tags
> ZIMP=false READLINE=false ZLIB=false GMP=false (exactly as the
> previous version, except for GMP).
>
> I have a Branch-and-Price code, with quite a few more features
> (pricer, branching rule, heuristics, cuts). To make my code working I
> just needed to change the functions SCIP_DECL_PRICERREDCOST() and
> SCIPaddCut() adding the new parameters. All the rest is exactly the
> same. The number of nodes and columns generated is much bigger, and
> the dual bound is converging very slowly. The time goes from less than
> a second to some few hundred seconds.
>
> Do you have any insight of this behaviour? or anything in particular I
> should check that is behaving differently in the new version?
>
> Thanks in advance!
>
> best,
>
> Alessia
>
>
> -- 
> Alessia Violin
> Service Graphes et Optimisation Mathématique (G.O.M.)
> Université Libre de Bruxelles
> C.P. 210/01
> Boulevard du Triomphe
> B-1050 BRUXELLES
> Tel: 02 650 58 80 - Fax: 02 650 59 70
> Email: aviolin at ulb.ac.be
> Webpage: http://homepages.ulb.ac.be/~aviolin/
>
>
> _______________________________________________
> 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/20140312/3f119b3d/attachment.html>