public override TAction DecideAction(MDPPerception <TState> perception)
{
CurrentState = perception.GetState();
CurrentReward = perception.GetReward();
if (this.StartingTrial())
{
TAction chosenAction = this.SelectRandomAction();
this.UpdateLearnerState(chosenAction);
return(PreviousAction);
}
if (MDP.IsTerminalState(CurrentState))
{
this.IncrementStateActionCount(PreviousState, PreviousAction);
this.UpdateQ(0.8);
PreviousAction = null;
PreviousState = null;
//TODO: make sure that this is ok value for what used to be null in java
previousReward = double.NegativeInfinity;
return(PreviousAction);
}
else
{
this.IncrementStateActionCount(PreviousState, PreviousAction);
TAction chosenAction = this.UpdateQ(0.8);
this.UpdateLearnerState(chosenAction);
return(PreviousAction);
}
}
public override TAction DecideAction(MDPPerception <TState> perception)
{
if (!(utilityFunction.HasUtilityFor(perception.GetState())))
{ // if
// perceptionState
// is
// new
utilityFunction.SetUtility(perception.GetState(), perception.GetReward());
MDP.SetReward(perception.GetState(), perception.GetReward());
}
if (!(PreviousState == null))
{
stateCount.IncrementFor(PreviousState);
utilityFunction = this.UpdateUtilityFunction(1.0);
}
if (MDP.IsTerminalState(CurrentState))
{
PreviousState = null;
PreviousAction = null;
//TODO: make sure that 0 is appropriate value for what used to be null in java
previousReward = 0;
}
else
{
PreviousState = CurrentState;
PreviousAction = policy.GetAction(CurrentState);
previousReward = CurrentReward;
}
return(PreviousAction);
}
public override TAction DecideAction(MDPPerception <TState> perception)
{
if (!(utilityFunction.HasUtilityFor(perception.GetState())))
{ // if
// perceptionState
// is
// new
utilityFunction.SetUtility(perception.GetState(), perception
.GetReward());
MDP.SetReward(perception.GetState(), perception.GetReward());
}
if (!(PreviousState == null))
{
if (nsa.ContainsKey(new Pair <TState, TAction>(
PreviousState, PreviousAction)))
{
nsa[new Pair <TState, TAction>(PreviousState, PreviousAction)] += 1;
}
else
{
nsa[new Pair <TState, TAction>(PreviousState, PreviousAction)] = 1.0;
}
if (nsasdash.ContainsKey(new MDPTransition <TState, TAction>(PreviousState, PreviousAction, CurrentState)))
{
nsasdash[new MDPTransition <TState, TAction>(PreviousState, PreviousAction, CurrentState)] += 1;
}
else
{
nsasdash[new MDPTransition <TState, TAction>(PreviousState, PreviousAction, CurrentState)] = 1.0;
}
foreach (MDPTransition <TState, TAction> transition in nsasdash.Keys)
{
if (nsasdash[transition] != 0.0)
{
double newValue = nsasdash[transition]
/ nsa[new Pair <TState, TAction>(
transition.GetInitialState(), transition.GetAction())];
MDP.SetTransitionProbability(transition, newValue);
}
}
IList <MDPTransition <TState, TAction> > validTransitions = MDP
.GetTransitionsWith(PreviousState, policy.GetAction(PreviousState));
utilityFunction = this.ValueDetermination(validTransitions, 1);
}
if (MDP.IsTerminalState(CurrentState))
{
PreviousState = null;
PreviousAction = null;
}
else
{
PreviousState = CurrentState;
PreviousAction = policy.GetAction(CurrentState);
}
return(PreviousAction);
}