[SCIP] Fscip run time

[Pedro Matias Carrancio]

Thorsten,thank you very much for you response.

What you say is true, but with harder problems the same thing happens.

Pyscipopt log:

  222s| 13100 |  4121 |  1221k|  93.0 |     0  |  72 | 920 | 186 | 981 |
756k|  0 |   0 |   0 | 5.367782e-04 | 5.083094e-04 |   5.60%|  47.53%
(node 13111) solution of LP 139849 not optimal (pfeas=1, dfeas=0) --
solving again with tighter feasibility tolerance

SCIP Status        : solving was interrupted [gap limit reached]
Solving Time (sec) : 222.53
Solving Nodes      : 13111
Primal Bound       : +5.08309438454586e-04 (48 solutions)
Dual Bound         : +5.36773734174404e-04
Gap                : 5.60 %


Fscip Log  (-sth 30)

 314          34120         505        30           0.0005           0.0005
     5.59%           0.0005      5.59%
      320          34127          37        23           0.0005
0.0005      5.59%           0.0005      5.59%
      325          34138          37        12           0.0005
0.0005      5.59%           0.0005      5.59%
SCIP Status        : solving was interrupted [given gap reached]
Total Time         : 328.70
  solving          : 328.70
  presolving       : 0.06 (included in solving)
B&B Tree           :
  nodes (total)    : 34149
Solution           :
  Solutions found  : 2
  Primal Bound     : +5.08310004133510e-04
  Dual Bound       : +5.36729879827307e-04
Gap                : 5.59105 %

I used this params in fscip (normal ramp-up)

RacingStatBranching = FALSE
RampUpPhaseProcess = 0

In racing ramp-up the time is greater.

Maybe I'm missing some additional configuration. I don´t know.

Thanks for you help.

Pedro Carrancio.






El mié, 17 ene 2024 a las 18:02, Thorsten Koch (<koch at zib.de>) escribió:

> Hi Pedro,
>
> this problem is too small and easy to profit from Fiber SCIP.
>
> You only have 1 or 3 nodes. fscip parallelizes on a node basis. There is
> nothing to parallelize.
>
> BR Thorsten
>
> Am 17.01.24 um 20:49 schrieb Pedro Matias Carrancio:
> > Hello Scip Team.
> >
> > I  solved a MINLP by running scip in python with Pyscipopt.
> >
> > It ran ok , I had optimal response in 25 seconds. I exported the model
> in attached file. (modelMM.cip)
> >
> > image.png
> >
> > Now, I´m trying to solve the same MINLP with fscip.
> >
> > It run well too, but it takes longer.
> >
> > -sth 1
> >
> > image.png
> >
> > image.png
> >
> > And the greater the number of threads, the greater the delay.
> >
> > -sth 3
> >
> > image.png
> >
> > image.png
> >
> >
> > It´s right? I expected that with more threads the time would improve and
> that with one thread the time would be similar to that of pysicpopt.
> >
> > Any help is appreciated.
> >
> > Thanks.
> >
> >
> > Pedro Carrancio.
> >
> >
> >
> > _______________________________________________
> > Scip mailing list
> > Scip at zib.de
> > https://listserv.zib.de/mailman/listinfo/scip
>
> --
> The important thing is not to stop questioning.
> Curiosity has its own reason for existing.          -- Albert Einstein
> ______________________________________________________________________
> Prof. Dr. Thorsten Koch /       Technische Universität Berlin
> www.zib.de/koch        /                   and
> koch at zib.de           /          Zuse Institute Berlin (ZIB)
> _____________________/       Takustraße 7, 14195 Berlin, Germany
>         and also                Phone +49-30-84185-213, Fax -269
> Kooperativer Bibliotheksverbund Berlin Brandenburg (KOBV)
> Forschungs- und Kompetenzzentrum Digitalisierung Berlin (digiS)
> ______________________________________________________________________
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20240119/52b590a4/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: modelMM.cip
Type: application/octet-stream
Size: 47132 bytes
Desc: not available
URL: <http://listserv.zib.de/pipermail/scip/attachments/20240119/52b590a4/attachment.obj>