[SCIP] SCIProwGetObjParallelism assert query
Marc Pfetsch
pfetsch at mathematik.tu-darmstadt.de
Fri Jun 23 18:16:01 CEST 2023
Hi James,
in exact arithmetic parallelism should be at most 1. If this is not the
case, this might signify an error. This is why there is an assert. Or
the problem might occur because of numerical problems (you numbers look
quite big). In any case, the issue will be ignored in optimized mode.
If you want to enable some additional checks, you can add "#define
SCIP_MORE_DEBUG" at the top of lp.c, recompile and run. SCIP will then
check whether the values use in the function below are reasonably close
to the ones that it would get by recomputing the values. Note that this
will slow down execution.
Best
Marc
On 23/06/2023 16:55, James Cussens wrote:
> Hi all,
>
> I'm getting the following assert error when running in OPT=dbg mode
> (everything else default compilation options):
>
> gobnilp: src/scip/lp.c:7820: SCIProwGetObjParallelism: Assertion
> `SCIPsetIsSumLE(set, parallelism, 1.0)' failed.
>
> The relevant code from lp.c is:
>
> parallelism = SCIPsetIsPositive(set, prod) ? REALABS(row->objprod) /
> SQRT(prod) : 0.0;
> assert(SCIPsetIsSumGE(set, parallelism, 0.0));
> assert(SCIPsetIsSumLE(set, parallelism, 1.0));
> parallelism = MIN(parallelism, 1.0);
> parallelism = MAX(parallelism, 0.0);
>
> If it is reasonable to assert that parallelism is <= 1.0, then why do we
> need the "parallelism = MIN(parallelism, 1.0);" line?
>
> Here are the values that are causing the assert error:
> [lp.c:7819] debug:
> |row->objprod|=5.66927e+07,prod=7.62443e+14,parallelism=2.05316
>
> James
>
> James Cussens
> Room MVB 3.26
> Dept of Computer Science, University of Bristol
> Phone: +44 (0)117 455 8723
> https://jcussens.github.io/ <https://jcussens.github.io/>
> Funded PhDs available in Bristol in the following areas: Data Science
> <http://www.bristol.ac.uk/cdt/compass/>, Interactive AI
> <http://www.bristol.ac.uk/cdt/interactive-ai/>, Cyber Security
> <http://www.bristol.ac.uk/cdt/cyber-security/> or Digital Health
> <http://www.bristol.ac.uk/cdt/digital-health/>.
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
More information about the Scip
mailing list