[SCIP] Separating all Cliques from the Clique table

[Leona Gottwald]

Hello Andreas,
in your settings it seems that you might disable presolving? Please
make sure your cliques are acutally added to the clique table first and
preslving is not skipped completely.
Regarding other settings that you can try there is the parameter
"presolving/clqtablefac" which is 2 by default. It limits the number of
nonzeros  relative to the problem nonzeros that are stored in the
clique table and could prevent some of your cliques to be stored.
Otherwise increasing the limits of the parameters for separation that
you already tried, e.g. "separating/clique/maxsepacuts" and the
orthogonality setting and maybe also try increasing
"separating/clique/cliquetablemem".
It still might happen that the cuts from the clique separator are not
selected because the cutselection prefers other cuts. The only way to
make sure that really all the clique cuts  are separated is, to do as
you suggested and add them all as constraints, possibly with
initial=FALSE.
You can also try out how setting "constraints/setppc/cliquelifting =
TRUE" affects your model. This is a more expensive presolve step that
can merge cliques into larger ones.
Leona
Am Donnerstag, den 20.08.2020, 11:58 +0000 schrieb Linß, Andreas:
> Dear scip community,
> I have a huge amount of possible clique cuts, which i have as set
> packing constraints in the model.
> 
> 
> 
> Now I have exactly these inequalities in my presolver as cliques to
> the clique graph (SCIPaddClique).
> 
> 
> 
> When separating using the Set-Packing Constraint handler, all cliques
> are found and now constraints is violated. If I comment out the
> addition of the Set-Packing Constraints, some cliques are separated.
> However, some of the cliques I add to the clique graph
>  are still severely violated with a sum value of 1.5.  The
> corresponding cut also has an efficiousness of
> 0.133631.
> 
> 
> 
> 
> 
> Is it possible to tell the clique separator to separate all possible
> clique-cuts, that are violated by the current solution?
> 
> My actual settings in scip-optsuite-7.0.1 are :
> 
> 
> 
> 
> 
> presolving/../maxrounds=0
> heuristics/../freq = -1
> 
> propagating/../maxprerounds = 0
> 
> propagating/../freq = -1
> separating/../freq = -1
> 
> separating/clique/priority = 10000000
> 
> separating/clique/freq = 1
> 
> separate/minorthoroot = 0.1
> 
> separating/clique/maxsepacuts = 1000000
> 
> separate/minefficacyroot = 0.0001
> 
> separating/maxstallrounds = -1
> 
> separate/poolfreq = 1
> 
> separating/clique/maxzeroextensions = -1
> 
> 
> 
> 
> 
> 
> As an extension of my question: 
> 
> is it perhaps more efficient to have the clique cuts in the model as
> Set-Packing Constraints (initial = false, separate = true), even if
> there are a lot of them, and only separate more specific cliques with
> the Clique generator?
> 
> 
> 
> 
> 
> 
> Many thanks for the support!
> 
> 
> 
> 
> 
> Kind regards
> Andreas Linß
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________Scip mailing 
> listScip 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/20200821/30177e9f/attachment.html>