protected override Reward Playout(WorldModel initialPlayoutState)
{
WorldModel worldModel = initialPlayoutState.GenerateChildWorldModel();
int depthCount = 0;
while (!worldModel.IsTerminal() && depthCount <= MaxPlayoutDepth)
{
Action[] actions = worldModel.GetExecutableActions();
Action biasedAction = actions.First();
foreach (Action action in actions)
{
if (action.GetHValue(worldModel) < biasedAction.GetHValue(worldModel))
{
biasedAction = action;
}
}
biasedAction.ApplyActionEffects(worldModel);
depthCount++;
}
if (depthCount > MaxPlayoutDepthReached)
{
base.MaxPlayoutDepthReached = depthCount;
}
return(new Reward()
{
Value = GetWorldModelScore(worldModel),
PlayerID = initialPlayoutState.GetNextPlayer(),
});
}
protected virtual Reward Playout(WorldModel initialPlayoutState)
{
Action[] actions = initialPlayoutState.GetExecutableActions();
float score = 0;
foreach (Action action in actions)
{
WorldModel worldModel = initialPlayoutState.GenerateChildWorldModel();
int depthCount = 0;
while (!worldModel.IsTerminal())
{
Action randomAction = actions[RandomGenerator.Next(actions.Length)];
randomAction.ApplyActionEffects(worldModel);
depthCount++;
}
if (depthCount > MaxPlayoutDepthReached)
{
MaxPlayoutDepthReached = depthCount;
}
score += worldModel.GetScore();
}
return(new Reward()
{
Value = score / actions.Length,
PlayerID = initialPlayoutState.GetNextPlayer(),
});
}
protected override Reward Playout(WorldModel initialPlayoutState)
{
//WorldModel newState = initialPlayoutState.GenerateChildWorldModel();
FutureStateWorldModel newState = new FutureStateWorldModel((FutureStateWorldModel)initialPlayoutState);
Reward reward = new Reward();
int numberOfIterations = 0;
while (!newState.IsTerminal() && numberOfIterations <= DEPTH_LIMIT)
{
Action[] possibleActions = newState.GetExecutableActions();
List <float> heuristics = new List <float>();
for (int i = 0; i < possibleActions.Length; i++)
{
heuristics.Add(possibleActions[i].GetHValue(newState));
}
int index = chooseAction(possibleActions, heuristics);
Action bestAction = possibleActions[index];
bestAction.ApplyActionEffects(newState);
newState.CalculateNextPlayer();
reward.PlayerID = newState.GetNextPlayer();
reward.Value = heuristics[index];
if (DEPTH_LIMIT > 0)
{
numberOfIterations++;
}
}
return(reward);
}
protected MCTSNode Expand(MCTSNode parent, Action action)
{
WorldModel newState = parent.State.GenerateChildWorldModel();
action.ApplyActionEffects(newState);
MCTSNode child = new MCTSNode(newState)
{
Action = action,
Parent = parent,
PlayerID = newState.GetNextPlayer(),
};
parent.ChildNodes.Add(child);
return(child);
}
protected int chooseAction(Action[] possibleActions, List <float> heuristics)
{
Action bestAction = null;
float maxHeuristic = -1.0f;
int best = 0;
for (int i = 0; i < possibleActions.Length; i++)
{
if (heuristics[i] > maxHeuristic)
{
maxHeuristic = heuristics[i];
bestAction = possibleActions[i];
best = i;
}
}
return(best);
}
protected MCTSNode Expand(MCTSNode parent, Action action)
{
WorldModel newModel = parent.State.GenerateChildWorldModel();
action.ApplyActionEffects(newModel);
newModel.CalculateNextPlayer();
MCTSNode node = new MCTSNode(newModel)
{
Action = action,
Parent = parent,
PlayerID = newModel.GetNextPlayer(),
Q = 0,
N = 0
};
parent.ChildNodes.Add(node);
return(node);
}
// Only worth running multiple playouts in case action is Sword Attack
protected virtual WorldModel StochasticPlayout(Action action, WorldModel currState)
{
if (action.Name.Equals("SwordAttack") && this.MaxPlayouts > 0)
{
WorldModel[] tests = new WorldModel[this.MaxPlayouts];
for (int i = 0; i < this.MaxPlayouts; i++)
{
tests[i] = currState.GenerateChildWorldModel();
action.ApplyActionEffects(tests[i]);
}
currState = AverageState(tests, (SwordAttack)action);
}
else
{
currState = currState.GenerateChildWorldModel();
action.ApplyActionEffects(currState);
}
return(currState);
}
protected MCTSNode Expand(MCTSNode parent, Action action)
{
//TODO: implement
throw new NotImplementedException();
}