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 override void ActionProbabilities(ReinforcementLearningState state, ReinforcementLearningQStore qFunction, out ReinforcementLearningAction[] actions, out float[] probabilities)
{
actions = state.GetActions();
probabilities = new float[actions.Length];
float maxq = float.MinValue;
for(int i=0; i<actions.Length; ++i)
{
float q = qFunction.Evaluate(state, actions[i]);
probabilities[i] = q;
if (q > maxq) maxq = q;
}
float sum = 0;
for (int i = 0; i < actions.Length; ++i)
{
probabilities[i] = (float)Math.Exp((probabilities[i] - maxq) / Temperature);
sum += probabilities[i];
}
for (int i = 0; i < actions.Length; ++i)
{
probabilities[i] /= sum;
}
}