Decision Optimization

Decision Optimization

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


#Analytics
#DecisionOptimization
#DecisionOptimization
 View Only

  • 1.  How to warm-start with partial solution in Java API

    Posted 05/26/20 04:10 PM
    Edited by System Admin 01/20/23 04:16 PM
    Hi,

    I'd like to warm-start a MILP with multiple partial solutions. I am presenting a MWE below.

    int numNodes = 1000;

IloIntVar[] x = new IloIntVar[numNodes];
IloIntVar[] y = new IloIntVar[numNodes];
  
for(int i=0; i< numNodes; i++) { 
  x[i] = cplex.boolVar("x_"+ i);   
  y[i] = cplex.boolVar("y_"+ i);
}

HashMap<Integer, Set<Integer>> candidateSolutions = new HashMap<Integer, Set<Integer>>();

Path myFilePath = Paths.get(DATADIR).resolve(Paths.get("solutions.txt"));    	
try {
     Files.lines(myFilePath).forEach(line -> {
     String[] columnValues = line.split("\\s+");
     Set<Integer> yVars= new HashSet<Integer>();
     for(int i =1 ; i < columnValues.length; i++) {
          yVars.add(Integer.valueOf(columnValues[i]));
      }
      candidateSolutions.put(Integer.valueOf(columnValues[0]),leaves );
       });
} catch (IOException e) {
      e.printStackTrace();
}

// Suppose I have two different partial solutions. candidateSolutions are: 
// 1) key = (x[1] = 1),  values = (y[10] = 1, y[62] =1) 
// 2) key = (x[34] = 1), values =  (y[23] = 1, y[456] = 1, y[92] =1)​

    Before I have used warm-start when I have the full solution with multi-dimensional variables. However, I could not get away with the scenario shown above. I was wondering if someone could help me with this.

    Here is a CPLEX source for warm-star in Java API.






    ------------------------------
    Ser Giovio
    ------------------------------
    #DecisionOptimization


  • 2.  RE: How to warm-start with partial solution in Java API
    Best Answer

    Posted 05/27/20 01:17 AM
    If I understand correctly then an entry a => { b, c } in your map represents a solution x[a]=1, y[b]=1, y[c]=1.

    You have to convert this into a flat array of variables and values and then use IloCplex.addMIPStart as in this untested code:
    for (Map.Entry<Integer,Set<Integer>> e : candidateSolution) {
  IloNumVar[] vars = new IloNumVar[1 + e.getValue().getSize()];
  double[] val = new double[vars.length];
  int idx = 0;
  vars[idx] = x[e.getKey()];
  vals[idx++] = 1;
  for (Integer yIdx : e.getValue()) {
    vars[idx] = y[yIdx];
    vals[idx++] = 1;
  }
  cplex.addMIPStart(vars, vals);
}​
    Note that you may want to use a TreeMap instead of a HashMap for candidateSolution or may want to sort the candidate solutions before adding them as MIP starts.
    The order of MIP starts makes a difference and so you may want to have something that is guaranteed to be the same in every run.

    ------------------------------
    Daniel Junglas
    ------------------------------



  • 3.  RE: How to warm-start with partial solution in Java API

    Posted 05/27/20 04:46 AM
    Hi @Daniel Junglas,

    You're right that ​about "If I understand correctly then an entry a => { b, c } in your map represents a solution x[a]=1, y[b]=1, y[c]=1."

    Thanks for the help!  I started using TreeMap<Integer, Set<Integer>> candidateSolutions= new TreeMap<Integer, Set<Integer>>(); as you suggested. According to my preliminary tests, it seems like everything is working fine.


    ------------------------------
    Ser Giovio
    ------------------------------

    Original Message