private void Update()
{
SetParameters();
dic.AddOrUpdate(new State(DistanceToObstacleX, DistanceToObstacleY, PositionX), nic);
}
public override void LearningAlgorithm()
{
SetParameters();
switch (state)
{
case (int)QLearningStateMachine.DoAction:
{
actualState = GetState();
if (!qLambdaTable.ContainsKey(actualState))
{
qLambdaTable.AddOrUpdate(actualState, new EQValues());
}
selectedActionNumber = GetRandomAction(actualState);
ExecuteAction(selectedActionNumber);
state = (int)QLearningStateMachine.Observe;
break;
}
case (int)QLearningStateMachine.Observe:
{
nextState = GetState();
DontAllowBlockBot();
if (actualState != nextState)
{
float maxQvalue;
if (qLambdaTable.ContainsKey(nextState))
{
maxQvalue = qLambdaTable[nextState].QValues.Max();
}
else
{
qLambdaTable.AddOrUpdate(nextState, new EQValues());
maxQvalue = 0;
}
if (!rewardIsReached)
{
Reward = (nextState.State1 - actualState.State1) * Rewards.ForPosition;
}
if (enemyIsReached)
{
Reward = Rewards.EnemyTouched;
}
float delta = Reward + GAMMA * maxQvalue - qLambdaTable[actualState].GetQValue(selectedActionNumber);
qLambdaTable[actualState].SetEValue(selectedActionNumber, 1);
if (!rewardIsReached)
{
foreach (var key in qLambdaTable.Keys)
{
for (int action = 0; action < qLambdaTable[key].QValues.Length; action++)
{
qLambdaTable[key].QValues[action] += (ALPHA * delta * qLambdaTable[key].EValues[action]);
}
qLambdaTable[key].updateEValues(LAMBDA, GAMMA);
}
}
if (TerminationStateIsReached())
{
ResetETable();
SaveQLambdaDataToFile();
}
state = (int)QLearningStateMachine.DoAction;
}
break;
}
}
}
