我在C ++中实现A* algorithm
来解决 n-puzzle问题。
我试图在this链接中实现伪代码
总成本(F = H + G)计算是“成本取决于错位瓦片的数量(启发式)+从初始状态(G)开始的步骤”。下面给出AStar
函数的算法。
问题是,我有一个无限循环的情况。我该如何解决这个问题?
PS:如果需要,我可以提供AStar
中使用的其他功能的实现。
任何帮助都将不胜感激。
void AStar(const int size, int** puzzle)
{
int moveCount = 0; // initialize G(n)
int**goalState = GoalState(size); // initialize and assign goal state
int openListIndex = 0; // initialize open list index
vector<node> openList; // initialize open list
vector<node> closedList; // initialize closed list
node startNode; // initialize start node
startNode.puzzleArray = puzzle; // assign start node's state
startNode.cost = moveCount + Heuristics(goalState,puzzle,size); // assign start node's cost
node goalNode; // initialize goal node
goalNode.puzzleArray = goalState; // assign goal node's state
openList.push_back(startNode); // push start node to the open list
while (!openList.empty()) // loop while open list is not empty
{
node currentNode = CalculateLowestCost(&openList, &closedList); // initialize current node which has the lowest cost, pop it from open list, push it to the closed list
int** currentState = currentNode.puzzleArray; // initialize and assign current state array
/*********************************************************************************************/
if (GoalCheck(goalState, currentState, size)) break; // GOAL CHECK//
/*********************************************************************************************/
vector<char> successorDirectionList = CalculateSuccessor(size, currentState); // initialize a char vector for the directions of the successors
int**successor; // initialize successor state
node successorNode; // initialize successor node
moveCount++; // advance G(n)
for (;!successorDirectionList.empty();) // loop over the successor list
{
char direction = successorDirectionList.back(); // take a direction from the list
successorDirectionList.pop_back(); // remove that direction from the list
successor = MoveBlank(currentState, size, direction); // assign successor state
successorNode.puzzleArray = successor; // assign successor node's state
successorNode.cost = moveCount + Heuristics(goalState,currentState,size); // assign successor node's cost
//vector<node> stateCheckList = openList; // copy the open list for the checking the nodes in that list
bool flagOpen = false;
bool flagClosed = false;
int locationOpen = -1;
int locationClosed = -1;
for (int i=0; i<openList.size(); i++)
{
int** existing = openList[i].puzzleArray;
int existingCost = openList[i].cost;
if (StateCheck(successor, existing, size))
{
locationOpen = i;
if (successorNode.cost > existingCost)
{
flagOpen = true;
break;
}
}
}
if (flagOpen) continue;
int** existingInOpen;
if(locationOpen != -1)
{
existingInOpen = openList[locationOpen].puzzleArray;
openList.erase(openList.begin()+locationOpen);
}
for (int i=0; i<closedList.size(); i++)
{
int** existing = closedList[i].puzzleArray;
int existingCost = closedList[i].cost;
if (StateCheck(successor, existing, size))
{
locationClosed = i;
if (successorNode.cost > existingCost)
{
flagClosed = true;
break;
}
}
}
if (flagClosed) continue;
int**existingInClosed;
if(locationClosed != -1)
{
existingInClosed = closedList[locationClosed].puzzleArray;
closedList.erase(closedList.begin()+locationClosed);
}
openList.push_back(successorNode);
}
}
}
答案 0 :(得分:1)
由于循环的可能性,即一系列移动会将您带回到您已经访问过的状态,因此检查重复状态(显然不是树搜索的问题)非常重要。我无法完全按照您的检查方式进行检查,但这可能是问题所在。在编写Haskell实现时,我遇到了类似的问题(详细信息here和here),这归结为处理探索集的问题。做对了,一切正常。 (获得4x4拼图的解决方案仍然是一个昙花一现的事情,特别是如果你从状态空间的目标很远的一个州开始,但这主要是由于A *的缺陷和天真的方式我们正在处理可能的行动。)
答案 1 :(得分:1)
您是否删除了从打开列表中选择的状态?这是一个非常简单且写得很好的C#代码,它可以帮助你:http://geekbrothers.org/index.php/categories/computer/12-solve-8-puzzle-with-a 并且A *会自动避免循环,因为它采取了你考虑的动作
答案 2 :(得分:0)
我还实现了一个n-puzzle(深度优先和*)并且它进入了无限循环。这是因为以下循环:向上,向左,向下,向右,向上,向左... 我真的不知道是否有办法防止这样的事情(我能做的最大办法就是通过记住最后一次移动来阻止左右循环......)
不知道这是否是导致你的循环的原因,最好的方法是在每次迭代时打印电路板以查看实际发生的情况;)