private qTurple GetBestTurple(IState s)
{
IEnumerable <qTurple> list = null;
if (QValues.TryGetValue(s, out list))
{
if (list.Count() == 0)
{
throw new NoOperatorFoundException();
}
qTurple bestTurple = list.ElementAt(0);
foreach (qTurple turple in list)
{
if (turple.Q > bestTurple.Q)
{
bestTurple = turple;
}
}
return(bestTurple);
}
else
{
AddNewState(s);
return(GetBestTurple(s));
}
}
private void AddNewState(IState state)
{
List <qTurple> list = new List <qTurple>();
foreach (IOperator op in Env.ApplicableOperators(state))
{
qTurple turple = new qTurple();
turple.Q = DefaultQ(state, op);
turple.Op = op;
list.Add(turple);
}
QValues.Add(state, list);
}
private void Train()
{
episodeCount += 1;
EpisodeReward = 0;
IState currentState = Start.State;
while (!currentState.Equals(Goal.State) && stopwatch.Elapsed.TotalMilliseconds < timeAvailable && !stopped)
{
IOperator op = trainingPolicy.action(currentState);
IOutcome outcome = Env.act(currentState, op);
EpisodeReward += outcome.Reward;
qTurple turple = getTurple(currentState, op);
if (!goalFound && outcome.State.Equals(Goal.State))
{
turple.Q += 100;
}
turple.Q = turple.Q + Alpha * (outcome.Reward + Gamma * getTurple(outcome.State, GetBestOp(outcome.State)).Q - turple.Q);
currentState = outcome.State;
}
EpisodeRewards.Add(EpisodeReward);
TimesList.Add(stopwatch.Elapsed.TotalMilliseconds);
if (currentState.Equals(Goal.State))
{
goalFound = true;
}
}