[SCIP] small gaps

[Stefan Vigerske]

Hi,

if the objective is a convex quadratic, then setting
constraints/quadratic/sepanlpmincont = 0 or
constraints/quadratic/gaugecuts = TRUE or
constraints/quadratic/projectedcuts = TRUE or
may be worth a try.

Stefan

On 1/30/20 6:10 AM, Marcus Daniels wrote:
> Hi all,
> 
> I don’t know if it is relevant, but this quadratic constraint (that is proliferating convex cuts) arises from a quadratic objective.    There were no quadratic constraints in the input problem.
> 
> Best,
> 
> Marcus
> 
> From: Scip <scip-bounces at zib.de> on behalf of Marcus Daniels <marcus at snoutfarm.com>
> Date: Tuesday, January 28, 2020 at 9:03 AM
> To: Gerald Gamrath <gamrath at zib.de>, "scip at zib.de" <scip at zib.de>
> Subject: Re: [SCIP] small gaps
> 
> Hi Gerald, all,
> 
> I tracked down the source of the cuts to this line in cons_quadratic.c:
> 
> SCIP_CALL( separatePoint(scip, conshdlr, conss, nconss, nusefulconss, sol, SCIPfeastol(scip), TRUE, &separateresult, &sepaefficacy) );
> 
> In my problem, I see efficacies far above SCIPfeastol (1e-6), even above 1e6.  the result can be tens of thousands of cuts that don't do much but slow progress to a crawl.   My solves to optimality are far faster with this efficacy cutoff much higher (1e6), and more like CPLEX solve times.
> 
> Thanks,
> 
> Marcus
> ________________________________
> From: Gerald Gamrath <gamrath at zib.de>
> Sent: Monday, January 27, 2020 12:56 PM
> To: Marcus Daniels <marcus at snoutfarm.com>; scip at zib.de <scip at zib.de>
> Subject: Re: [SCIP] small gaps
> 
> Dear Marcus,
> 
> which SCIP version are you using?
> 
> There is probably some branching and cuts going on, but SCIP does not print information for every node. But it seems indeed very slow here. If you interrupt using CTRL-C or set a time limit of say, one hour, and then print the statistics via "display statistics" you should be able to identify where all the time is spent. Alternatively, you could also "set display frequency 1" to print a line for every node. But in the end, the statistics should be most interesting.
> 
> Best,
> Gerald
> On 27.01.20 18:59, Marcus Daniels wrote:
> Hi,
> 
> I have a set of relaxed problems that find feasibility, run for a few seconds, and then get stuck as below.   There's no further branching and no further cuts, not even after an hour.    Depending on the problem the gap might be 0.01% or 0.80%.   Any suggestions on how to debug this or tune it to get to a zero gap?  I tried doing less frequent LP solves, but that also runs forever.
> 
> Thanks,
> 
> Marcus
> 
>   time | node  | left  |LP iter|LP it/n| mem |mdpt |frac |vars |cons |cols |rows |cuts |confs|strbr|  dualbound   | primalbound  |  gap
>   97.1s|     1 |     0 |  1247 |     - | 156M|   0 |   2 |5502 |3048 |5502 |3067 |  20 |   0 |   0 | 1.524548e-02 | 1.528577e-02 |   0.26%
>   97.1s|     1 |     2 |  1247 |     - | 157M|   0 |   2 |5502 |3048 |5502 |3067 |  20 |   0 |   2 | 1.524548e-02 | 1.528577e-02 |   0.26%
> 
> 
> 
> 
> _______________________________________________
> 
> Scip mailing list
> 
> Scip at zib.de<mailto:Scip at zib.de>
> 
> https://listserv.zib.de/mailman/listinfo/scip
> 
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>