[Cbc] Troubles between cbc and cplex
Mathieu LACROIX
lacroix at isima.fr
Tue Jul 8 03:25:10 EDT 2008
Hi,
I'm trying to use Cbc with cplex and I have a problem.
When I use Clp for the linear solver, my branch and cut algorithm is ok.
But when I change the linear solver to cplex
(OsiSolverInterface * solver = new OsiCpxSolverInterface instead of
OsiClpSolverInterface), I have some troubles. The linear relaxation is
solved without any problem... The branch-and-cut is then applied, the
optimal value is found but automatically, the solution is claimed
infeasible (which is not the case with Clp). I report the few last lines
of the optimization
Cbc0031I 6 added rows had average density of 31.5
Cbc0013I At root node, 6 cuts changed objective from 228 to 454 in 7 passes
Cbc0014I Cut generator 0 (VULNERABILITE) - 6 row cuts (6 active), 0 column cuts in 0.012 seconds - new frequency is 0
Cbc0023I Allowing solution, even though largest row infeasibility is 1
Cbc0021I On closer inspection node is infeasible
Cbc0001I Search completed - best objective 1e+50, took 31 iterations and 0 nodes (0.02 seconds)
Cbc0035I Maximum depth 0, 0 variables fixed on reduced cost
(In this example, there exists a feasible solutin with objective value
equal to 454). How can I specify to cplex that the solution is feasible?
I also have another problem. For an instance, the linear relaxation took
nearly one hour to be solved with clp. When I replace clp by cplex, it
takes half an hour. However, when I take the mps file and solve it with
cplex in interactive use, it takes less than 5 minutes. The difference
may came from the parameters. How can I speed up the linear relaxation's
time execution?
Best regards,
Mathieu
--
Mathieu LACROIX
mathieu.lacroix at isima.fr
Université Blaise Pascal - Clermont Fd II
Laboratoire LIMOS
Bâtiment ISIMA - Bureau D112
Complexe scientifique des Cézeaux
63 177 Aubière, Cedex - France
Tel : 04.73.40.79.48
Fax : 04.73.40.76.39
More information about the Cbc
mailing list