OptaPlaner simple example cant find feasible solution - optaplanner

Question

to get familiar with optaplanner i created a simple test project. I only have one Solution and one Entity class. The Entity has only one value between 0 and 9. There should only be odd numbers and the sum of all should be less then 10 (this are just some random constraints i came up with).
As Score i use a simple HardSoftScore. Here is the code:
public class TestScoreCalculator implements EasyScoreCalculator<TestSolution>{
#Override
public HardSoftScore calculateScore(TestSolution sol) {
int hardScore = 0;
int softScore = 0;
int valueSum = 0;
for (TestEntity entity : sol.getTestEntities()) {
valueSum += entity.getValue() == null? 0 : entity.getValue();
}
// hard Score
for (TestEntity entity : sol.getTestEntities()) {
if(entity.getValue() == null || entity.getValue() % 2 == 0)
hardScore -= 1; // constraint: only odd numbers
}
if(valueSum > 10)
hardScore -= 2; // constraint: sum should be less than 11
// soft Score
softScore = valueSum; // maximize
return HardSoftScore.valueOf(hardScore, softScore);
}
}
and this is my config file:
<?xml version="1.0" encoding="UTF-8"?>
<solver>
<!-- Domain model configuration -->
<scanAnnotatedClasses/>
<!-- Score configuration -->
<scoreDirectorFactory>
<easyScoreCalculatorClass>score.TestScoreCalculator</easyScoreCalculatorClass>
</scoreDirectorFactory>
<!-- Optimization algorithms configuration -->
<termination>
<secondsSpentLimit>30</secondsSpentLimit>
</termination>
</solver>
for some reason OptaPlanner cant find a feasible solution. It terminates with LS step (161217), time spent (29910), score (-2hard/10soft), best score (-2hard/10soft)... and the solution 9 1 0 0.
So the hardScore is -2 because the two 0 are not odd. A possible solution would be 7 1 1 1 for example. Why is this ? This should be a really easy example ...
(when i set the Start values to 7 1 1 1 it terminates with this solution and a score of (0hard/10soft) how it should be)
Edit:
The Entity class
#PlanningEntity
public class TestEntity {
private Integer value;
#PlanningVariable(valueRangeProviderRefs = {"TestEntityValueRange"})
public Integer getValue() {
return value;
}
public void setValue(Integer value) {
this.value = value;
}
#ValueRangeProvider(id = "TestEntityValueRange")
public CountableValueRange<Integer> getStartPeriodRange() {
return ValueRangeFactory.createIntValueRange(0, 10);
}
}
The Solution class
#PlanningSolution
public class TestSolution {
private List<TestEntity> TestEntities;
private HardSoftScore score;
#PlanningEntityCollectionProperty
public List<TestEntity> getTestEntities() {
return TestEntities;
}
public void setTestEntities(List<TestEntity> testEntities) {
TestEntities = testEntities;
}
#PlanningScore
public HardSoftScore getScore() {
return score;
}
public void setScore(HardSoftScore score) {
this.score = score;
}
#Override
public String toString() {
String str = "";
for (TestEntity testEntity : TestEntities)
str += testEntity.getValue()+" ";
return str;
}
}
The Main Program class
public class Main {
public static final String SOLVER_CONFIG = "score/TestConfig.xml";
public static int printCount = 0;
public static void main(String[] args) {
init();
}
private static void init() {
SolverFactory<TestSolution> solverFactory = SolverFactory.createFromXmlResource(SOLVER_CONFIG);
Solver<TestSolution> solver = solverFactory.buildSolver();
TestSolution model = new TestSolution();
List<TestEntity> list = new ArrayList<TestEntity>();
// list.add(new TestEntity(){{setValue(7);}});
// list.add(new TestEntity(){{setValue(1);}});
// list.add(new TestEntity(){{setValue(1);}});
// list.add(new TestEntity(){{setValue(1);}});
for (int i = 0; i < 4; i++) {
list.add(new TestEntity());
}
model.setTestEntities(list);
// Solve the problem
TestSolution solution = solver.solve(model);
// Display the result
System.out.println(solution);
}
}

Answer 1

It gets stuck in a local optima because there is no move that takes 1 from entity and gives it to another entity. With a custom move you can add that.
These kind of moves only apply to numeric value ranges (which are rare, usually value ranges are a list of employees etc), but they should probably exist out of the box (feel free to create a jira for them).
Anyway, another way to get the good solution is to add <exhaustiveSearch/>, that bypassing local search and therefore the local optima. But that doesn't scale well.