[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Help-glpk] suboptimal solution from glp_intopt
From: |
Andrew Makhorin |
Subject: |
Re: [Help-glpk] suboptimal solution from glp_intopt |
Date: |
Fri, 25 Oct 2013 19:08:53 +0400 |
On Fri, 2013-10-25 at 15:41 +0200, Jonas Behr wrote:
> Hi,
>
> I would like to switch from CPLEX to glpk and encountered the following
> problem.
>
> The solution returned by glp_intopt satisfies all constraints, but when
> compared to the
> CPLEX solution it is clearly suboptimal. If I however add additional equality
> constraints
> that fix all variables to values of the CPLEX solution, then glpk returns the
> correct solution
> and also reports that this solution is optimal. So I can see that the optimal
> solution is in the
> feasible set the way I specified the problem for glpk.
>
> Am I missing something? Is there any additional information needed?
> Please find the output below.
>
> Best regards,
> Jonas
>
>
> // output from CPLEX
> Solution status = Optimal
> Solution value = 834.531
>
>
> //solve lp using glpk
> //num_constraints: 2038, cnt:11100
> 0: obj = 0.000000000e+00 infeas = 2.472e+01 (198)
> 200: obj = 4.570010471e+01 infeas = 7.385e+00 (112)
> * 266: obj = 1.562382132e+02 infeas = 0.000e+00 (54)
> * 317: obj = 1.672094849e+01 infeas = 0.000e+00 (54)
> OPTIMAL SOLUTION FOUND
> glpk simplex obj: 16.721
> ipp_basic_tech: 1187 row(s) and 106 column(s) removed
> ipp_reduce_bnds: 2 pass(es) made, 86 bound(s) reduced
> ipp_basic_tech: 221 row(s) and 32 column(s) removed
> ipp_reduce_coef: 3 pass(es) made, 31 coefficient(s) reduced
> glp_intopt: presolved MIP has 630 rows, 503 columns, 1947 non-zeros
> glp_intopt: 17 integer columns, all of which are binary
> Scaling...
> A: min|aij| = 5.000e-01 max|aij| = 1.000e+00 ratio = 2.000e+00
> Problem data seem to be well scaled
> Constructing initial basis...
> Size of triangular part = 630
> Solving LP relaxation...
> 317: obj = -4.009109495e+02 infeas = 4.385e+00 (0)
> 400: obj = 9.121388715e+02 infeas = 1.282e+00 (0)
> * 413: obj = 9.367043712e+02 infeas = 0.000e+00 (0)
> * 514: obj = 5.912600642e+02 infeas = 0.000e+00 (0)
> OPTIMAL SOLUTION FOUND
> Integer optimization begins...
> Cover cuts enabled
> + 514: mip = not found yet >= -inf (1; 0)
> + 532: >>>>> 9.385649858e+02 >= 9.363705197e+02 0.2% (7; 0)
> + 537: >>>>> 9.367043712e+02 >= 9.367043712e+02 0.0% (5; 2)
> + 537: mip = 9.367043712e+02 >= tree is empty 0.0% (0; 13)
> INTEGER OPTIMAL SOLUTION FOUND
> //glpk mip obj: 936.704
> //my obj is: 936.704
>
>
Not seeing your mip it is difficult to say what is wrong. Please write
your mip in mps or cplex lp format (or better in glpk format if you are
using glpk api), gzip it, and post it to me (not to the list). Thanks.
Andrew Makhorin
- Re: [Help-glpk] suboptimal solution from glp_intopt,
Andrew Makhorin <=