public override ReinforcementLearningAction ActionForState(ReinforcementLearningState state, ReinforcementLearningQStore qFunction)
{
Random r = new Random();
if (r.NextDouble() > EPSILON)
{
return new InvertedPendulumAction2((float)r.NextDouble());
}
else
{
ReinforcementLearningAction ret;
float value;
qFunction.GetBestActionAndUtilityForState(state, out ret, out value);
return ret;
}
}
public override void ActionProbabilities(ReinforcementLearningState state, ReinforcementLearningQStore qFunction, out ReinforcementLearningAction[] actions, out float[] probabilities)
{
ReinforcementLearningAction action;
float utility;
actions = state.GetActions();
probabilities = new float[actions.Length];
qFunction.GetBestActionAndUtilityForState(state, out action, out utility);
for (int i = 0; i < actions.Length; ++i)
{
if (actions[i].Equals(action))
{
probabilities[i] = Epsilon;
}
else probabilities[i] = (1 - Epsilon) / (actions.Length - 1);
}
}
public new ReinforcementLearningAction ActionForState(ReinforcementLearningState state, ReinforcementLearningQStore qFunction)
{
ReinforcementLearningAction action;
float utility;
qFunction.GetBestActionAndUtilityForState(state, out action, out utility);
return action;
}