Originally posted by: abhzap

Hi All,

     In the final step of CG, we do the LP to MIP conversion and solve the master problem (MIP version).

     However, I am not able to change my model variables from float to int.

     Here is how I go about doing it (looking at cutstock. java example)

      //convert to an integer version of the problem

             //converting the X variables
             for ( Map.Entry<MpVarX, IloNumVar> var : usedVarX.entrySet() ) {                
                 master_problem.add(master_problem.conversion(var.getValue(), IloNumVarType.Int));          
             }

             //converting the Y variables
             for ( Map.Entry<MpVarY, IloNumVar> var : usedVarY.entrySet() ) {                 
                 master_problem.add(master_problem.conversion(var.getValue(), IloNumVarType.Int));
             }

             //converting the Z variables
             for ( Map.Entry<MpVarZ, IloNumVar> var : usedVarZ.entrySet()  ) {
                 master_problem.add(master_problem.conversion(var.getValue(), IloNumVarType.Int));            
             }

             I check the type of the variables after solving the master problem and find that it is still float!

             Is there a issue with the way I am using Java?

-abhzap

Assuming the map values are the IloNumVars in your master problem, this looks correct to me. Have you tried either tracing execution in a debugger or sticking print statements in the loops to verify that the calls to IloCplex.add() are actually occurring (with the expected arguments)?

Originally posted by: abhzap

Hi Paul,

     Thank you for the reply. I checked and it seems the IloConversion object is being added. For e.g. the code below

           //converting the X variables
             for ( Map.Entry<MpVarX, IloNumVar> var : usedVarX.entrySet() ) {                
                 master_problem.add(master_problem.conversion(var.getValue(), IloNumVarType.Int)); 
                 IloConversion convObj = master_problem.conversion(var.getValue(), IloNumVarType.Int);
                 master_problem.add(convObj);                  
             }

         gives the error -> CPLEX Error: multiple type conversions on a variable' caught; so the IloCoversion object is being added to the model.

          Also, when write the solution using : master_problem.writeSolution("jsol"); The information included there actually shows the model is solved as an MIP.  A part of the information is shown below

 <?xml version = "1.0" standalone="yes"?>

<CPLEXSolution version="1.2">

 <header
   problemName="ilog.cplex"
   solutionName="incumbent"
   solutionIndex="-1"
   objectiveValue="7000"
   solutionTypeValue="3"
   solutionTypeString="primal"
   solutionStatusValue="101"
   solutionStatusString="integer optimal solution"
   solutionMethodString="mip"
   primalFeasible="1"
   dualFeasible="1"
   MIPNodes="0"
   MIPIterations="0"
   writeLevel="1"/>

 <quality
   epInt="1e-05"
   epRHS="1e-06"
   maxIntInfeas="1.00000010050394e-07"
   maxPrimalInfeas="0"
   maxX="1"
   maxSlack="9999998"/>

           The solution is however still like below:

 X_1 = 1.0
 X_2 = 1.00000010000001E-7
 X_3 = 1.00000010000001E-7
 X_4 = 1.00000010000001E-7
 X_5 = 1.00000010000001E-7
 Y_1 = 1.00000010000001E-7
 Y_2 =  1.00000010000001E-7
 Y_3 = 1.0
 Y_4 = 0.99999989999999
 Y_5 = 0.99999989999999

So the integer conversion actually is working. All those values are within rounding tolerance of integer. The default integrality tolerance for CPLEX (the IloCplex.Param.MIP.Tolerances.Integrality parameter) is 1E-5, and all those values are well within 1e-5 of either 0 or 1.

Originally posted by: abhzap

Thanks for the help Paul! This resolves one of my issues.

Originally posted by: TOYSED

Hello,

I also have a problem with the use of IloConversion. I have a Column Generation code. At the last step, I convert the parameters of my MasterModel to boolean with IloConversion but the code could not compute the minimum value.

here is the conversion;

IloModel IntegerMaster(env);
 IloCplex cplex1(IntegerMaster);
 IntegerMaster.add(Mmodel);
 
 for (j = 0; j <nofG; j++)
  IntegerMaster.add(IloConversion (env, P[j], ILOBOOL));
 
 cplex1.solve();
 cplex1.exportModel("d://IntegerMasterModel.LP");
 cout<<"Objective ="<< cplex1.getObjValue()<<endl;

And the resultant model;

Minimize
 obj: 538 x1 + 509 x2 + 242 x3 + 338 x4 + 89 x5 + 35 x6 + 146 x7 + 155 x8
Subject To
 c1: x1 + x6 + x7 - x9  = 0
 c2: x2 + x8 - x10  = 0
 c3: x3 + x7 + x8 - x11  = 0
 c4: x4 - x12  = 0
 c5: x5 + x6 - x13  = 0
 c6: x1 + x2 + x3 + x4 + x5 + 5 x6 + 5 x7 + 5 x8  = 2
Bounds
 0 <= x1 <= 1
 0 <= x2 <= 1
 0 <= x3 <= 1
 0 <= x4 <= 1
 0 <= x5 <= 1
 0 <= x6 <= 1
 0 <= x7 <= 1
 0 <= x8 <= 1
 0 <= x9 <= 1
 0 <= x10 <= 1
 0 <= x11 <= 1
 0 <= x12 <= 1
 0 <= x13 <= 1
Binaries
 x1  x2  x3  x4  x5  x6  x7  x8
End

Original models 

What do you mean by "could not compute minimum value"? Can you please elaborate?

When I put your model into an LP file then it solves here in a blink.

Originally posted by: TOYSED

Thanks for your fast response. 

When I solve this model, I obtain an objective function value of 331 where x3 and x5 are 1. But this solution is not the minimum . I expect to obtain a solution of 190.

 It seems like that the code only considers the first five variables which are there before column generation. Although I see all the variables in the final model 

the following code does not consider variables x6, x7 and x8 added to the Master Model. Thus I obtain an objective function value of 331 instead of 190. I wonder if this is related to wrong use of IloConversion or not? How can I solve this issue?

Originally posted by: TOYSED

I figured out my problem...I taught it is related with IloConversion because I focused on that. Thanks...:)