[SCIP] Running SCIP with branching disabled

[Vladimir V. Voloshinov]

Dear Mark,
on June, 14, my colleague, Sergey Smirnov, asked you about the issue:
"How to set SCIP parameters so that the solver will use only cuts and will
not use branching?"
We tried your recommendation (see below) but did not get success...

Recently we returned to our "research on cuts-only-and-no-branching effect"
with the latest version of SCIP 7.0.3.
We try PySCIPOpt now and the method
model.writeParams(onlychanged=True)
write the following non-default settings
limits/totalnodes = 1
separating/maxcuts = 10000000
separating/maxcutsroot = 10000000
separating/cutagelimit = -1
branching/random/priority = 20000

But in the output I get the *[total node limit reached]*:
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
....
presolved problem has 2633 variables (2633 bin, 0 int, 0 impl, 0 cont) and
1888 constraints
     16 constraints of type <setppc>
      2 constraints of type <xor>
    804 constraints of type <linear>
    395 constraints of type <orbitope>
    671 constraints of type <logicor>
transformed objective value is always integral (scale: 1)
Presolving Time: 0.16
....
 time | node  | left  |LP iter|LP it/n|mem/heur|mdpt |vars |cons |rows
|cuts |sepa|confs|strbr|  dualbound   | primalbound  |  gap   | compl.
  0.9s|     1 |     0 |   579 |     - |    47M |   0 |2379 |1758 |1381 |
0 |  0 |  10 |   0 | 0.000000e+00 |      --      |    Inf | unknown
  0.8s|     1 |     0 |  1087 |     - |    52M |   0 |2402 |1768 |1524 |
132 |  1 |   9 |   0 | 0.000000e+00 |      --      |    Inf | unknown
.....
  4.7s|     1 |     2 | 13383 |     - |    66M |   0 |2379 |1778 |2840
|1459 | 11 |  10 |   0 | 0.000000e+00 |      --      |    Inf | unknown

SCIP Status        : solving was interrupted [total node limit reached]
Solving Time (sec) : 4.68
Solving Nodes      : 1
Primal Bound       : +1.00000000000000e+20 (0 solutions)
Dual Bound         : +0.00000000000000e+00
Gap                : infinite
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
More over, it seems that all parameters except limits/totalnodes , do not
take any effect on the result!?

After some "investigation" I tried to set the following parameters also (in
addition to those mentioned above)
constraints/linear/sepafreq = 20000
constraints/linear/propfreq = 20000
constraints/linear/proptiming = 15

And for the one (only one!) instance of our problem (they are generated
"randomly" from some predefined class) , the solving process took about 70
seconds and the number of cuts reached ~7000...
But I can not reproduce this effect for other  instances of our problem !!!
Can you help us with proper settings of SCIP parameters to tell the solver
use only cuts and do not use branching?

Sincerely yours,
Vladimir V. Voloshinov,
Ph.D, GoogleScholar profile
<https://scholar.google.ru/citations?hl=en&user=-m4QhNEAAAAJ&view_op=list_works&sortby=pubdate>

On Mon, Jun 14, 2021 at 8:51 PM Marc Pfetsch <
pfetsch at mathematik.tu-darmstadt.de> wrote:

>
>
> Dear Sergey,
>
> you need to add a node limit and turn off strong-branching. In the
> console this can be done as follows:
>
> set limits nodes 1
> set branching random priority 20000
>
> (The last command prioritizes the random branching rule of the default
> rule with strong branching).
>
> Strong branching might also be the reason for the observed behavior.
>
> Best
>
> Marc
>
>
> On 14/06/2021 11.57, Sergey Smirnov wrote:
> > Dear SCIP community,
> > can you tell me if it is possible to run SCIP in such a way, when only
> > cuts (at root node) will be used and no branching?
> > We use SCIP in "a console mode" as a standalone application.
> >
> > My question is triggered by an effect my colleague found: one of
> > commercial solvers solves some of Integer Linear Programming problems
> > much faster by cuts only, without any branching. It was a kind of
> > Constrained Programming problem, to find a feasible solution of a system
> > of linear equations with binary variables.
> > "Only cuts" took a few seconds and "cuts and branching" - a few minutes.
> >
> > Regards,
> > Sergey
> >
> >
> > _______________________________________________
> > Scip mailing list
> > 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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20211013/b7c88079/attachment.html>