[SCIP] Question about "type" of MIPLIB instances
Gregor Hendel
hendel at zib.de
Mon Aug 26 09:31:39 CEST 2019
Hi Tony,
the type classification uses the constraints statistics of the problem
after presolving of the final run, e.g., after the last restart. The
most likely explanation is that conflict analysis created bound
disjunction constraints, which are not linearly representable. If this
happens during the first run, and then SCIP performs a restart, such
constraints become a permanent part of the problem, and the type
classification says "CIP", which is the most general problem class it
knows. You can find the type classification code in check/check.awk.
The statistics may look as follows:
Constraints : Number MaxNumber #Separate #Propagate #EnfoLP #EnfoRelax #EnfoPS #Check #ResProp Cutoffs DomReds Cuts Applied Conss Children
benderslp : 0 0 0 0 2 0 0 18 0 0 0 0 0 0 0
integral : 0 0 0 0 2 0 0 18 0 0 0 0 0 0 4
varbound : 93 93 12 1257 0 0 0 0 0 0 0 0 0 0 0
setppc : 37 37 35 4937 0 0 0 11 8 2 0 0 0 0 0
linear : 655 655 35 4940 0 0 0 5 34 56 40 0 0 0 0
orbitope : 1 1 0 3989 0 0 0 4 0 0 0 0 0 0 0
logicor : 5 5 35 77 0 0 0 0 0 0 0 0 0 0 0
bounddisjunction : 2 2 0 177 0 0 0 0 0 0 0 0 0 0 0
Besides bound disjunction constraints, the second explanation could be
the extraction of auxiliary and constraints within the gate extraction
presolver.
Best,
Gregor
Am 24.08.19 um 01:52 schrieb Wang, Tony Congqian:
>
> Hello community,
>
>
> I have used SCIP on the MIPLIB2017 benchmark set with make test. At
> the end I noticed that some instances where classified as type "CIP"
> although every Problem was obviously a MIP. Does this happen when
> logic or constraints are used by scip? Does this actually make a
> difference in how the programm works compared to the MIP instances?
>
>
> Best regards and many thanks,
>
> Tony
>
>
> _______________________________________________
> 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/20190826/a2578bb9/attachment.html>
More information about the Scip
mailing list