为什么我的AVL树排序算法花费的时间比插入排序要长?

时间:2017-04-21 16:43:21

标签: c++ algorithm sorting binary-search-tree avl-tree

我正在研究一个AVL树排序算法,我想我终于弄明白了,感谢大家的帮助,直到我意识到它的运行时间明显长于运行时间插入排序,这不应该是正确的。我使用随机生成数字的未排序数组(或更确切地说,向量)。我将在下面提供一些统计数据和代码。

AVL

for (std::vector<int>::const_iterator i = numbers.begin(); i != numbers.begin()+30000; ++i)
{
    root = insert(root, numbers[x]);
    cout << "Height: " << height(root);
    x++;
    track++;
    if( (track % 10000) == 0)
    {
        cout << track << " iterations" << endl;
        time_t now = time(0);
        cout << now - begin << " seconds" << endl;
    }

}

N = 30,000

身高= 17

执行的迭代次数= ~1,730,000

运行时间排序= 38秒

插入排序

for (int i = 0; i < 30000; i++)
    {
        first++;
        cout << first << " first level iterations" << endl;
        time_t now = time(0);
        cout << now - begin << " seconds" << endl;
        int tmp = dataSet[i];
        int j;
        for (j = i; i > 0 && tmp < dataSet[j - 1]; j--)
        {
            dataSet[j] = dataSet[j - 1];
        }
        dataSet[j] = tmp;
    }
}

N = 30,000

迭代次数= 30,000

运行时间排序= 4秒

这可能是正确的,所以我希望你们都可以帮助弄清楚发生了什么?据我所知,我的所有代码都已正确实现,但我仍然会在下面包含相关部分,以防错过。

源代码

    node* newNode(int element) // helper function to return a new node with empty subtrees
{
    node* newPtr = new node;
    newPtr->data = element;
    newPtr->leftChild = NULL;
    newPtr->rightChild = NULL;
    newPtr->height = 1;
    return newPtr;
}
node* rightRotate(node* p) // function to right rotate a tree rooted at p
{
    node* child = p->leftChild;
    node* grandChild = child->rightChild;

    // perform the rotation
    child->rightChild = p;
    p->leftChild = grandChild;

    // update the height for the nodes
    p->height = max(height(p->leftChild), height(p->rightChild)) + 1;
    child->height = max(height(child->leftChild), height(child->rightChild)) + 1;

    // return new root
    return child;
}
node* leftRotate(node* p) // function to left rotate a tree rooted at p
{
    node* child = p->rightChild;
    node* grandChild = child->leftChild;

    // perform the rotation
    child->leftChild = p;
    p->rightChild = grandChild;

    // update heights
    p->height = max(height(p->leftChild), height(p->rightChild)) + 1;

    // return new root
    return child;
}

int getBalance(node *p)
{
    if(p == NULL)
        return 0;
    else
        return height(p->leftChild) - height(p->rightChild);
}
// recursive version of BST insert to insert the element in a sub tree rooted with root
// which returns new root of subtree
node* insert(node*& n, int element)
{
    // perform the normal BST insertion
    if(n == NULL) // if the tree is empty
        return(newNode(element));
    if(element< n->data)
    {
        n->leftChild = insert(n->leftChild, element);
    }
    else
    {
        n->rightChild = insert(n->rightChild, element);
    }

    // update the height for this node
    n->height = 1 + max(height(n->leftChild), height(n->rightChild));

    // get the balance factor to see if the tree is unbalanced
    int balance = getBalance(n);

    // the tree is unbalanced, there are 4 different types of rotation to make
    // Single Right Rotation (Left Left Case)
    if(balance > 1 && element < n->leftChild->data)
    {
        return rightRotate(n);
    }
    // Single Left Rotation (Right Right Case)
    if(balance < -1 && element > n->rightChild->data)
    {
        return leftRotate(n);
    }
    // Left Right Rotation
    if(balance > 1 && element > n->leftChild->data)
    {
        n->leftChild = leftRotate(n->leftChild);
        return rightRotate(n);
    }
    // Right Left Rotation
    if(balance < -1 && element < n->rightChild->data)
    {
        n->rightChild = rightRotate(n->rightChild);
        return leftRotate(n);
    }
    // cout << "Height: " << n->height << endl;
    // return the unmodified root pointer in the case that the tree does not become unbalanced
    return n;
}

1 个答案:

答案 0 :(得分:1)

“调试”此类问题的最佳方法是use a performance profiler tool,但在这种情况下,我认为我可以给你一个很好的假设:

总之,您的比较并非“公平”。

如何“修复”这个?:

例如,如果您事先知道树将拥有多少个节点,则可以在算法开始时一次分配所有必需的节点,并使用它们创建节点池(如果使用Boost应该比每次调用标准new运算符快得多,一个接一个)。

每次需要新节点时,都可以从池中获取。然后,您可以从不需要额外内存分配的点“比较”算法。