/**
* prints solution after maximizing v without violating constraints for debugging purpose
*
* @param solution = the solution
* @param nrVar = the number of variables in the solution
*/
private void printSolution(RealPointValuePair solution, int nrVar) {
for (int i = 0; i < nrVar; i++) {
System.out.print("pi" + i + "=" + solution.getPoint()[i] + " ");
}
System.out.println(" V = " + solution.getValue());
System.out.println(" V = " + solution.getPoint()[nrVar] + "\n\n");
}
/**
* Maximize V while not violating the constraints
*
* @param state = state the maximization has to take place for
* @return = array with values for pi's and v
* @throws OptimizationException
*/
private double[] solveEquations(int state) throws OptimizationException {
Collection constraints = new ArrayList();
// for each possible action of the prey
for (int preyAction = 0; preyAction < Action.nrActions; preyAction++) {
// initialize weigths for this constraint
double[] Q = new double[Action.nrActions + 1];
// for each possible action of the predator
for (int predAction = 0; predAction < Action.nrActions; predAction++) {
int newStatePred = policy.getLinearIndexForAction(state, Action.getAction(predAction));
int newStatePrey =
policy.getLinearIndexForAction(newStatePred, Action.getReverseAction(preyAction));
// calculate expected reward R(s,a,o)
double expReward = 0;
if (preyAction == Action.Wait.getIntValue()) {
expReward = policy.getReward(newStatePrey, false);
} else {
expReward =
policy.getReward(newStatePrey, false) * (1.0 - Ptrip)
+ policy.getReward(newStatePred, false) * Ptrip;
}
// add weight to constraint for this combitnation
if (preyAction == Action.Wait.getIntValue()) {
Q[predAction] = expReward + learningRate * vValues.getV(newStatePrey);
} else {
Q[predAction] =
expReward
+ learningRate * vValues.getV(newStatePrey) * (1.0 - Ptrip)
+ learningRate * vValues.getV(newStatePred) * Ptrip;
}
}
// add constraint weight for V
Q[Action.nrActions] = -1.0;
// add constraint
constraints.add(new LinearConstraint(Q, Relationship.GEQ, 0));
}
// add constraints that probabilities need to be > 0
for (int predAction = 0; predAction < Action.nrActions; predAction++) {
double[] constraintProb = new double[Action.nrActions + 1];
Arrays.fill(constraintProb, 0.0);
constraintProb[predAction] = 1.0;
constraints.add(new LinearConstraint(constraintProb, Relationship.GEQ, 0));
}
// add total is zero constraint
double[] totalZero = new double[Action.nrActions + 1];
Arrays.fill(totalZero, 1.0);
totalZero[Action.nrActions] = 0.0;
constraints.add(new LinearConstraint(totalZero, Relationship.EQ, 1.0));
// build objective function
double[] objective = new double[Action.nrActions + 1];
Arrays.fill(objective, 0.0);
objective[Action.nrActions] = 1.0;
LinearObjectiveFunction f = new LinearObjectiveFunction(objective, 0);
// solve and return
RealPointValuePair solution =
new SimplexSolver().optimize(f, constraints, GoalType.MAXIMIZE, false);
return solution.getPoint();
}
