Originally posted by: Chathura Thilanka

Hello,

I am trying to do a warm start using the Java API and having some issues when passing the initial solution to the model.

In my model file(.mod) I have a 2D decision variable defined as,

range nodes = 1..5;

range vehicles = 1..2;

dvar int service_time[nodes][vehicles];

 

In my java file I am building the model as below and trying to pass the initial solution to the above decision variable,

     //adding the custom data source

 

   //generating the model

      opl.generate();

 

//creating the initial solution

 

 

 

// adding the MIPStart

 

if(cplex.solve())  {

  // more code

}else{

 // more code

}

 // more code

   }catch(Exception ex){

        // more code

    }

}

 

Unfortunately I am having an exception in the line which calls the cplex.addMIPStart() function as 

     [java] ### CONCERT exception: The referenced IloExtractable has not been extracted by the IloAlgorithm
     [java] ilog.concert.IloException: The referenced IloExtractable has not been extracted by the IloAlgorithm
     [java]     at ilog.cplex.cppimpl.cplex_wrapJNI.IloCplex_addMIPStart__SWIG_0(Native Method)
     [java]     at ilog.cplex.cppimpl.IloCplex.addMIPStart(IloCplex.java:866)
     [java]     at ilog.cplex.IloCplex.addMIPStart(IloCplex.java:13219)
     [java]     at ilog.cplex.IloCplex.addMIPStart(IloCplex.java:13228)
     [java]     at myJavaClass.myJavaClass.main(myJavaClass.java:412)

 

I am thinking the error is due to the way I prepare the initial solution, can somebody please help me to sort this out.

 

Thank you very much.

 

#DecisionOptimization
#OPLusingCPLEXOptimizer

hi

have you had a look at the example at

CPLEX_Studio128\opl\examples\opl\warmstart

?

regards

#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Chathura Thilanka

Thanks Alex for the reply.

Yes, I looked at that example too. There they have not used the Java API to generate and solve the model. The issue is with passing the initial solution to the model through Java by addMIPStart() method. I was unable to find any example regarding this.

 

Thanks.

#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi,

let me help you to bridge that gap:

1) addMIPStart with 2D

range r = 1..10;
dvar int+ x[r][r];
dvar int+ y[r][r];
// The following array of values (defined as data) will be used as
// a starting solution to warm-start the CPLEX search.
float values[i in r][j in r] = ((i==5) &&(j==5))? 10 : 0;

minimize
  sum( i in r ) x[i][i] + sum( j in r ) y[j][j];
subject to{
  ctSum:    
    sum( i in r ) x[i][i] >= 10;
  forall( j in r )
    ctEqual:
      y[j][j] == j;
      
      forall(i,j in r : i != j)
        x[i][j] == 0;
      
}

main{
  thisOplModel.generate();  
  var def = thisOplModel.modelDefinition;   
  // Default behaviour
  writeln("Default Behaviour");
  var cplex1 = new IloCplex();
  var opl1 = new IloOplModel(def, cplex1);
  opl1.generate();
  cplex1.solve();   
  writeln(opl1.printSolution());
  // Setting initial solution
  writeln("Setting initial solution");
  var cplex2 = new IloCplex();
  var opl2 = new IloOplModel(def, cplex2);
  opl2.generate();
 
  cplex2.addMIPStart(opl2.x,opl2.values) ;
 
  cplex2.solve();   
  writeln(opl2.printSolution());

  opl1.end();
  cplex1.end();
  opl2.end();
  cplex2.end();
  0;
}

2) Examples on how to call OPL from java at CPLEX_Studio128\opl\examples\opl_interfaces\java

In mulprod_main you will notice

dataElements.getElement("Capacity").asNumMap().set("flour",capFlour);

In your

cplex.addMIPStart(flat_var2D, flat_var2D_startingVals, IloCplex.MIPStartEffort.Auto, "addMIPStart start");

you should have for flat_var2D a 2D array that was extracted and that is in the cplex matrix

regards

#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Chathura Thilanka

Thank you very much for your answer. It was helpful.

I could successfully perform a warm start for my model as follows,

There are 2 decisions variables in my model as,

int x = 1..10; int y = 1..5;

I created 2 vectors in my mod file as,

and filled them with a feasible initial solution for a warm start.

 

Then in my java code I attached the 2 initial solution vectors to the decision variables as,

 

By this way I could successfully do a warm start. I am wondering whether vecs.setStart() functions performs a  warm start same as in cplex.addMIPStart() function since the documentation says what setStart() does is it "restores the MIP start information".

 

Thank You.

 

 

#DecisionOptimization
#OPLusingCPLEXOptimizer

The difference between vecs.setStart() and cplex.addMIPStart() is the following:

setStart() is the preferred function when you use OPL. It goes through the OPL layer and eventually sets this single start as MIP start for the underlying IloCplex object.

addMIPStart() adds a MIP start to the underlying IloCplex object behind OPL's back. That is in general not a problem but not exactly nice. Additionally, as the method names suggest, using addMIPStart() you can add more than one MIP start while with setStart() you can only set one.

BTW, here is the reason why your first attempt with addMIPStart() did not work: You created new variables and used those to define a MIP start. There is no relation between those variables and the variables defined in your .mod file. The right thing to do here would be to obtain the existing variables from the IloOplModel instance and use those to define the MIP start. It is a lot either to just do everything in the .mod, though.

#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Chathura Thilanka

Thank you very much for your explanation.

Yeah, initially I was doing the addMIPStart() in a wrong way as you mentioned. The issue I had was extracting the decision variables from the model in as a IloNumVar[] to input to the  addMIPStart() function, as it is expecting IloNumVar[] type of an array as var array.

 

For example when I try to do the following in my Java code,

    IloIntRange nodes = opl.getElement("nodes").asIntRange();
    IloIntRange vehicles = opl.getElement("vehicles").asIntRange();
    IloNumVarMap serviceTime = opl.getElement("service_time").asNumVarMap();

    final int nbNodes = nodes.getSize();
    final int nbVehicles = vehicles.getSize();

    IloNumVar[] startX = new IloNumVar[nbNodes * nbVehicles];
    double[] startVals = new double[nbNodes * nbVehicles];
    for (int i = 0; i < nbNodes; i++) {
       IloNumVarMap inner = serviceTime.getSub(nodes.getValue(i));
       for (int j = 0; j < nbVehicles; j++) {
          int idx = i * nbVehicles + j;
          startX[idx] = inner.get(vehicles.getValue(j));
          startVals[idx] = 1.0;
       }
    }

    cplex.addMIPStart(startX, startVals);

 

I am getting an exception as,

### CONCERT exception: Type dvar int[nodes][vehicles] expected for element "service_time". 
[java] at ilog.opl.IloOplElement.cpp_asNumVarMap(IloOplElement.java:506)

 

Thank you.

#DecisionOptimization
#OPLusingCPLEXOptimizer

This sort of introspection is always a little complicated (which probably means you should avoid it ).

First of all, since your variables are of type 'int' you need to convert to IntVarMap instead of NumVarMap. But then you have to jump through some more hoops. Here is a code that seems to work for your model (no guarantees):

IloIntVarMap serviceTime = opl.getElement("service_time").asIntVarMap();
System.out.println("Dimensions: " + serviceTime.getNbDim());
// Since the ranges used to index the variable arrays in the .mod file may not be
// 0-based or may be sparse we need indexer objects to access them.
IloDiscreteDataCollectionArray indexer = serviceTime.makeMapIndexer();    
IloDiscreteDataCollection firstIndex = indexer.get(0);
IloDiscreteDataCollection secondIndex = indexer.get(1);
System.out.println(firstIndex.getName() + ": " + firstIndex + ", " + firstIndex.isIntRange());
System.out.println(secondIndex.getName() + ": " + secondIndex + ", " + secondIndex.isIntRange());
IloIntVar[][] service_time = new IloIntVar[firstIndex.getSize()][];
IloIntRange firstRange = (IloIntRange)firstIndex;
IloIntRange secondRange = (IloIntRange)secondIndex;
for (int i = 0; i < firstIndex.getSize(); ++i) {
  IloIntVarMap sub = serviceTime.getSub(i + firstRange.getLB());
  service_time[i] = new IloIntVar[sub.getSize()];
  for (int j = 0; j < secondIndex.getSize(); ++j) {
    IloIntVar v = sub.get(j + secondRange.getLB());
    service_time[i][j] = v;
    System.out.print(v);
  }
  System.out.println();
}

 

#DecisionOptimization
#OPLusingCPLEXOptimizer

similar questions at https://stackoverflow.com/questions/48662074/cplex-warm-start-error-when-using-opl-model-in-java-api/48673431#48673431

#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Chathura Thilanka

Hi Daniel,

Thanks for your response, I tried the conversion from NumVarMap  to IntVarMap. The issue I am facing is  addMIPStart() expecting (IloNumVar[] vars,double[] values). Since my variables are of int type I am stuck in the addMIPStart() calling due to the argument mismatch.

 

Thank you.

#DecisionOptimization
#OPLusingCPLEXOptimizer

You have to copy the two-dimensional variable/value arrays into a single flat array.

#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Chathura Thilanka

Thanks Daniel, It worked!

#DecisionOptimization
#OPLusingCPLEXOptimizer