public double update(RealVector x_t, RealVector x_tp1, double r_tp1, double gamma_tp1) {
if (x_t == null) return initEpisode();
v_t = v.dotProduct(x_t);
delta_t = r_tp1 + gamma_tp1 * v.dotProduct(x_tp1) - v_t;
v.addToSelf(alpha_v * delta_t, x_t);
return delta_t;
}
public double update(RealVector x_t, Action a_t, RealVector x_tp1, Action a_tp1, double r_tp1) {
if (x_t == null) return initEpisode();
Action atp1_star = greedy.decide(x_tp1);
RealVector phi_sa_t = toStateAction.stateAction(x_t, a_t);
delta = r_tp1 + gamma * greedy.bestActionValue() - theta.dotProduct(phi_sa_t);
if (a_t == at_star) e.update(gamma * lambda, phi_sa_t);
else {
e.clear();
e.update(0, phi_sa_t);
}
theta.addToSelf(alpha * delta, e.vect());
at_star = atp1_star;
return delta;
}
@Override
public double predict(RealVector phi) {
return v.dotProduct(phi);
}
@Override
public void resetWeight(int index) {
theta.setEntry(index, 0);
}
@Override
public double predict(RealVector phi_sa) {
return theta.dotProduct(phi_sa);
}