[Scip] scaling

[Tobias Achterberg]

Hi Tamas,

I think you should first check whether Soplex is the issue. For this, you can just link 
SCIP to CPLEX instead of Soplex. Just go into the "scip/" directory and enter

make LPS=cpx

Then, you will be asked for the path to the CPLEX include files, the libcplex.a, and a 
libcplex.so file. If you have the .a file, you don't need the .so file and can just hit 
enter for the .so file question.

After having compiled SCIP/CPLEX, just try it. If it finds a solution, the issue is in 
Soplex. If it doesn't find a solution, the issue is in SCIP's presolving. In both cases, 
it would be nice if would report your findings to the scip mailing list.


Best, Tobias


Tamas Mahr wrote:
> Hi all,
> 
> I am trying to solve a MIP using ziboptsuite-1.1.0. I have just
> downloaded and compiled it. It contains scip-1.1.0 and soplex-1.4.0.
> 
> If I try to solve the attached mip problem by the command
> scip-1.1.0/bin/scip -c "read dvrp.mps" -c "optimize", it comes up
> infeasible. I know that solutions to the problem exist, the cplex solver
> solves it in no time.
> 
> If I make the solver more verbose:
> scip-1.1.0/bin/scip -c "set display verblevel 5" -c "read dvrp.mps" -c
> "optimize", I get the following output:
> 
> SCIP version 1.1.0 [precision: 8 byte] [memory: block] [mode:
> optimized] [LP solver: SoPlex 1.4.0]
> Copyright (c) 2002-2008 Konrad-Zuse-Zentrum fuer Informationstechnik
> Berlin (ZIB)
> 
> user parameter file <scip.set> not found - using default parameters
> 
> parameter <display/verblevel> set to 5
> original problem has 11090 variables (11025 bin, 0 int, 0 impl, 65
> cont) and 4630 constraints
> 
> LP Solver <SoPlex 1.4.0>: dual feasibility tolerance cannot be set --
> tolerance of SCIP and LP solver may differ
> LP Solver <SoPlex 1.4.0>: barrier convergence tolerance cannot be set
> -- tolerance of SCIP and LP solver may differ
> LP Solver <SoPlex 1.4.0>: fastmip setting not available -- SCIP
> parameter has no effect
> LP Solver <SoPlex 1.4.0>: scaling not available -- SCIP parameter has
> no effect
> LP Solver <SoPlex 1.4.0>: presolving not available -- SCIP parameter
> has no effect
> transformed problem has 11090 variables (11025 bin, 0 int, 0 impl, 65
> cont) and 4630 constraints
>    4630 constraints of type <linear>
> 
> presolving:
> (round 1) 0 del vars, 195 del conss, 130 chg bounds, 0 chg sides, 0 chg
> coeffs, 0 upgd conss, 0 impls, 210 clqs
> (round 2) 1259 del vars, 1611 del conss, 1389 chg bounds, 3016 chg
> sides, 3846 chg coeffs, 0 upgd conss, 0 impls, 210 clqs
> (round 3) 1259 del vars, 1611 del conss, 1389 chg bounds, 3016 chg
> sides, 3846 chg coeffs, 210 upgd conss, 0 impls, 210 clqs
> ...
> SCIP Status        : problem is solved [infeasible]
> Solving Time (sec) : 22.01
> Solving Nodes      : 1
> Primal Bound       : +1.00000000000000e+20 (0 solutions)
> Dual Bound         : +1.00000000000000e+20
> Gap                : 0.00 %
> 
> In the beginning of the output you can see that SoPlex is complaining
> about scaling and presolving not being available features! The problem
> contains large numbers, I think it concludes infeasibility due to
> rounding errors. Thus scaling would be crucial here.
> 
> I have looked into the code, and scaling and presolving is certainly
> implemented in soplex, only the scip-soplex interface class does not
> make any use of them. I have even tried to modify this interface class,
> so it could find a feasible initial solution, but during the
> branch-and-bound the program died. I think it is because the BB
> algorithm assumes an unscaled, unsimplified problem in the soplex
> solver, which is not the case after my modifications.
> 
> Can you tell me, what am I doing wrong?
> 
> Tamas
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip