/**
* The policy is learned here
*
* @param sweeps = number of sweeps executed
* @throws OptimizationException
*/
public void learn(int sweeps) throws OptimizationException {
largestDif = 0.0;
// temporary table to store new V-values
StateRepV newV = new StateRepV(0.0, false);
for (int i = 0; i < sweeps; i++) {
double diff = 0.0;
// for each state
for (int state = 0; state < StateRepV.nrStates; state++) {
// solve the set of equations
double[] values = solveEquations(state);
// calculate difference in v-value
if (state != 0) {
diff = Math.abs(vValues.getV(state) - values[Action.nrActions]);
if (diff > largestDif) {
largestDif = diff;
}
newV.setValue(state, values[Action.nrActions]);
}
// repair values returned by solving the equations if neccessary
for (int a = 0; a < Action.nrActions; a++) {
if (values[a] < 0.00000001) {
policy.setValue(state, Action.getAction(a), 0.00000001);
} else if (values[a] > 0.99999999) {
policy.setValue(state, Action.getAction(a), 1.0);
} else {
policy.setValue(state, Action.getAction(a), values[a]);
}
}
}
// put new values in V-table
vValues = newV;
newV = new StateRepV(init, false);
}
}
