Decision Optimization

Decision Optimization

Delivers prescriptive analytics capabilities and decision intelligence to improve decision-making.

 View Only
  • 1.  feasible problem cannot be solved with mps

    Posted Thu December 31, 2015 05:26 PM

    Originally posted by: PingLiu


    I have a mixed integer linear problem (MILP). The problem has zero columns. It succeeded when solved with Matlab/Cplex interface.

     [x, fval, exitflag, output] = cplexmilp (double(f'), double(Aineq), double(bineq'), double(Aeq), double(beq'),...
          [ ], [ ], [ ], double(lb'), double(ub'), ctype, [ ], []);

    But when I tried to solve it in NEOS with the mps file, it failed. I am not sure why the same problem can not be solved in mps format.


    #CPLEXOptimizers
    #DecisionOptimization


  • 2.  Re: feasible problem cannot be solved with mps

    Posted Mon January 04, 2016 01:40 AM

    How did you create the Model.mps file? The file is invalid. In the NEOS output you can see

    CPLEX Error  1449: Line 913: 'C1' is not a column name.
    No file read.

    So CPLEX cannot even read your MPS file. That is why it cannot solve anything.


    #CPLEXOptimizers
    #DecisionOptimization


  • 3.  Re: feasible problem cannot be solved with mps

    Posted Mon January 04, 2016 02:45 PM

    Originally posted by: PingLiu


    I created the mps file (fullDense.mps in the attachement) through OPenopt/lpsolve interface. when I removed the zero column, then Cplex can correctly solve the reduced problem as attached in redDenseMPS.mps which is created in the same way as the mps before removing zero columns.

    A strange thing I found is that the fullDense.mps cannot be solved by either CPLEX or CBC or GLPK, but it can be successfully solved by SCIP with the same results as solved with matrix. SO I don't think the mps file is not properly created.

    Do you have any idea why this would happen?

    Thanks a lot.


    #CPLEXOptimizers
    #DecisionOptimization


  • 4.  Re: feasible problem cannot be solved with mps

    Posted Tue January 05, 2016 02:50 AM

    IMO the MPS file is wrong: it specifies an upper bound for a column that does not exist. Apparently, CPLEX, CBC, and GLPK correct detect that and fail. Other solvers (SCIP?) seem to silently ignore this error. I am not convinced that silently ignoring this error is a good idea.


    #CPLEXOptimizers
    #DecisionOptimization


  • 5.  Re: feasible problem cannot be solved with mps

    Posted Tue January 05, 2016 02:02 PM

    Originally posted by: PingLiu


    If you think silent ignoring is the reason for the mps error, why would it succeed when solving the MIP problem with matrix (e.g. Aeq, AIneq, beq, bineq, etc.) in Matlab/CPlex with a column which does not exist? Does it ignore this error silently too?


    #CPLEXOptimizers
    #DecisionOptimization


  • 6.  Re: feasible problem cannot be solved with mps

    Posted Wed January 06, 2016 01:51 AM

    Sorry, I don't understand what you are saying. You said that SCIP can read the broken MPS file, fine. But how is the CPLEX matlab interface related to that?


    #CPLEXOptimizers
    #DecisionOptimization


  • 7.  Re: feasible problem cannot be solved with mps

    Posted Wed January 06, 2016 01:26 PM

    Originally posted by: PingLiu


    the Matlab/cplex  accepts the matrix as inputs. Even though the matrix includes columns which do not exist, the Matlab/cplex interface can still successfully solve it. The input matrix is corresponding to the mps we talked about which contains nonexistent columns. Since the Matlab/zcplex interface can work on the matrix with nonexistent columns, so it means that Matlab/Cplex interface also ignore the nonexistent columns silently. Do I understand it correctly?


    #CPLEXOptimizers
    #DecisionOptimization