0
私は現在、単純なボードゲームのためのMCTSの実装に取り組んでいます。 私はそれがほぼ正しいと思うが、1回以上反復したい場合は、 （時間または数字（i）を反復の要因とする）、ゲームがフリーズする。誰かがその問題が何であるかについて確固たるアイディアを持っていれば嬉しいです。モンテカルロツリーの検索
// method to find best turn
Node findBest()
{
Node startingPoint = new Node();
startingPoint.CurrentField_Boxes = GameBoard.boxs;
startingPoint.CurrentField_H = GameBoard.horizontal_line;
startingPoint.CurrentField_V = GameBoard.vertical_line;
Node rootNode = startingPoint;
for(int i = 0; i < 2; i++)
{
//Selection
Node promisingNode = selectPromisingNode(rootNode);
Debug.Log("POST SELECTION CHECK VON X/Y " + promisingNode.x + "/" + promisingNode.y);
//Expansion
if (!checkField())
{
expandNode(promisingNode);
}
//Simulation
Node nodeToExplore = promisingNode;
Debug.Log("POST 2 SELECTION CHECK VON X/Y " + nodeToExplore.x + "/" + nodeToExplore.y);
if (promisingNode.getChildren().Count > 0)
{
//get random child node
nodeToExplore = promisingNode.getRandomChild();
Debug.Log("POST 3 SELECTION CHECK VON X/Y " + nodeToExplore.x + "/" + nodeToExplore.y);
}
int playoutResult = simulation(nodeToExplore);
Debug.Log("SIM RESULT :" + playoutResult);
//backpropagation
backPropogation(nodeToExplore);
}
Debug.Log("*********************************");
Node bestNode = new Node();
Debug.Log("POST 4 SELECTION CHECK VON X/Y " + rootNode.getBestChild().x + "/" + rootNode.getBestChild().y);
bestNode = rootNode.getBestChild();
return bestNode;
}
private Node selectPromisingNode(Node rootNode)
{
Debug.Log("SELECTING");
Node node = new Node();
node = rootNode;
while (node.getChildren().Count != 0) // checking if terminal
{
findBestNodeWithUCT(node);
}
return node;
}
Node findBestNodeWithUCT(Node root)
{
double j = 0;
int bestIndex = 0; ;
for (int i = 0; i < root.getChildren().Count; i++)
{
if (root.getChildren()[i].getVisitTimes()/root.getChildren()[i].getResult() > j)
{
j = root.getChildren()[i].getVisitTimes()/root.getChildren()[i].getResult();
bestIndex = i;
Debug.Log("BEST INDEX : " + bestIndex);
}
}
Node bestOption = root.getChildren()[bestIndex];
Debug.Log("BEST OPTION RETURN X/Y : " + bestOption.x + "/" + bestOption.y);
return bestOption;
}
private void expandNode(Node node)
{
Debug.Log("Start EXPANSION");
bool vert = false;
bool hori = false;
//Phase I : creating new Node
//create a new node wit certain action A
Node newNode = new Node();
//init and setting parent
newNode.parent[0] = node;
//init gamefield into node
newNode.CurrentField_V = node.CurrentField_V;
newNode.CurrentField_H = node.CurrentField_H;
//init and setting child parent relationship
newNode.visitTimes++;
Debug.Log("ERSTELLE NEUEN KNOTEN");
//Phase II : Validating Action
//create a action a which leads to the expansion
int x = UnityEngine.Random.Range(0,4);
int y = UnityEngine.Random.Range(0, 4);
//checking variables for valid coordinates
Debug.Log("AKTION A IN EXPANSION :" + x + y);
//Prechecking of generated coordinates to set corresponding obj
if (y < 3)
{
vert = true;
}
if (x < 3)
{
hori = true;
}
else if (x == 3 && y == 3)
{
Debug.Log(" OUT OF BOUNDS :NEW RANDOM");
expandNode(node);
}
//Phase III : Doing Action
if(vert && newNode.CurrentField_V[x, y].tag == "is play" && hori && newNode.CurrentField_H[x, y].tag != "is play")
{
expandNode(node);
}
//setting choice into sim gamefield with green color
if (vert && newNode.CurrentField_V[x, y].tag != "is play")// && GameBoard.sim_vertical_line[x, y].tag != "is play")
{
Debug.Log("SETZEN DER VERTIKALEN AKTION IN KNOTENSTATE");
GameBoard.sim_vertical_line[x, y].tag = "is play";
GameBoard.sim_vertical_line[x, y].GetComponent<SpriteRenderer>().sprite = greenV;
// newNode.CurrentField_V[x, y].tag = "is play";
newNode.x = x;
newNode.y = y;
node.nodeChildren.Add(newNode);
}
if (hori && newNode.CurrentField_H[x, y].tag != "is play")//&& GameBoard.sim_horizontal_line[x, y].tag != "is play")
{
Debug.Log("SETZEN DER HORIZONTALEN AKTION IN KNOTENSTATE");
GameBoard.sim_horizontal_line[x, y].tag = "is play";
GameBoard.sim_horizontal_line[x, y].GetComponent<SpriteRenderer>().sprite = greenH;
// newNode.CurrentField_H[x, y].tag = "is play";
newNode.x = x;
newNode.y = y;
node.nodeChildren.Add(newNode);
}
else { expandNode(node); }
}
private void backPropogation(Node nodeToExplore)
{
Node tempNode = nodeToExplore;
Debug.Log("BACKPROPAGATION");
while (tempNode != null)
{
int i = 0;
if (i % 2 == 0)
{
tempNode.visitTimes++;
int j = tempNode.result;
tempNode.setResult(j);
tempNode = tempNode.getParent();
}
else if (i % 2 != 0)
{
tempNode.visitTimes++;
int j = tempNode.result;
tempNode.setResult(j * -1);
tempNode = tempNode.getParent();
}
i++;
}
}
一回の反復が正常に動作し、私はすでにランダムボットのより良い振る舞いを得たが、たったの約5-6ノード最大で1回の反復で作成されますすることができます。
ありがとうございます。