protected int maxValue(State s, int alpha, int beta, int counter) {
// System.out.println(counter);
if (s.isTerminal() || counter > maxdepth) return utility(s);
else if (!s.isTerminal()) {
List<Move> moves = s.getPossibleMoves();
int maxv = Integer.MIN_VALUE;
// System.out.println(moves.size());
for (Move move : moves) {
// System.out.println("max");
int v = minValue(result(s, move), alpha, beta, counter + 1); // utility(s);
/*int m = minValue(result(s,move),alpha,beta,counter++);
if( m > v)
v = m;*/
// System.out.println();
if (v > maxv) {
maxv = v;
}
if (maxv >= beta) return maxv;
alpha = Math.max(alpha, maxv);
}
return maxv;
} else return 0;
}
public static float trajectoryReward(
State initialState,
Policy p,
int nbOfSteps,
LinkedList<State> statesTrajectory,
LinkedList<Action> actionsTrajectory,
LinkedList<Float> rewardsTrajectory) {
State s = initialState;
Action a = p.chooseAction(s);
float trajectoryReward = reward(s, a);
if (statesTrajectory != null) {
statesTrajectory.add(s);
actionsTrajectory.add(a);
rewardsTrajectory.add(reward(s, a));
}
int nbOfIterations = 1;
while (!s.isTerminal() && nbOfIterations < nbOfSteps) {
s = MarkovDecisionProcess.nextState(s, a);
a = p.chooseAction(s);
trajectoryReward += reward(s, a);
if (statesTrajectory != null) {
statesTrajectory.add(s);
actionsTrajectory.add(a);
rewardsTrajectory.add(reward(s, a));
}
nbOfIterations++;
}
return trajectoryReward;
}
protected int minValue(State s, int alpha, int beta, int counter) {
// System.out.println(counter);
if (s.isTerminal() || counter > maxdepth) return utility(s);
else if (!s.isTerminal()) {
List<Move> moves = s.getPossibleMoves();
int minv = Integer.MAX_VALUE;
// System.out.println(moves.size());
for (Move move : moves) {
// System.out.println("min");
int v = maxValue(result(s, move), alpha, beta, counter + 1); // utility(s);
/*int m = maxValue(result(s,move),alpha,beta,counter++);
if( m < v)
v = m;*/
if (v < minv) {
minv = v;
}
if (minv <= alpha) return minv;
beta = Math.min(beta, minv);
}
return minv;
} else return 0;
}
public static float reward(State s, Action a) {
float reward;
if (s.isTerminal()) {
reward = 0f;
} else {
State nextState = MarkovDecisionProcess.nextState(s, a, false);
float nextStepDistance = nextState.getRelativeDistance();
if (nextStepDistance < SoccerParams.KICKABLE_MARGIN) {
reward = 1000f;
} else {
reward = -1f;
}
}
return reward;
}
