637 lines
15 KiB
Markdown
637 lines
15 KiB
Markdown
# AVL 树
|
||
|
||
在「二叉搜索树」章节中提到,在进行多次插入与删除操作后,二叉搜索树可能会退化为链表。此时所有操作的时间复杂度都会由 $O(\log n)$ 劣化至 $O(n)$ 。
|
||
|
||
如下图所示,执行两步删除结点后,该二叉搜索树就会退化为链表。
|
||
|
||
![degradation_from_removing_node](avl_tree.assets/degradation_from_removing_node.png)
|
||
|
||
再比如,在以下完美二叉树中插入两个结点后,树严重向左偏斜,查找操作的时间复杂度也随之发生劣化。
|
||
|
||
![degradation_from_inserting_node](avl_tree.assets/degradation_from_inserting_node.png)
|
||
|
||
G. M. Adelson-Velsky 和 E. M. Landis 在其 1962 年发表的论文 "An algorithm for the organization of information" 中提出了「AVL 树」。**论文中描述了一系列操作,使得在不断添加与删除结点后,AVL 树仍然不会发生退化**,进而使得各种操作的时间复杂度均能保持在 $O(\log n)$ 级别。
|
||
|
||
换言之,在频繁增删查改的使用场景中,AVL 树可始终保持很高的数据增删查改效率,具有很好的应用价值。
|
||
|
||
## AVL 树常见术语
|
||
|
||
「AVL 树」既是「二叉搜索树」又是「平衡二叉树」,同时满足这两种二叉树的所有性质,因此又被称为「平衡二叉搜索树」。
|
||
|
||
### 结点高度
|
||
|
||
在 AVL 树的操作中,需要获取结点「高度 Height」,所以给 AVL 树的结点类添加 `height` 变量。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* AVL 树结点类 */
|
||
class TreeNode {
|
||
public int val; // 结点值
|
||
public int height; // 结点高度
|
||
public TreeNode left; // 左子结点
|
||
public TreeNode right; // 右子结点
|
||
public TreeNode(int x) { val = x; }
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
「结点高度」是最远叶结点到该结点的距离,即走过的「边」的数量。需要特别注意,**叶结点的高度为 0 ,空结点的高度为 -1** 。我们封装两个工具函数,分别用于获取与更新结点的高度。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 获取结点高度 */
|
||
int height(TreeNode node) {
|
||
// 空结点高度为 -1 ,叶结点高度为 0
|
||
return node == null ? -1 : node.height;
|
||
}
|
||
|
||
/* 更新结点高度 */
|
||
void updateHeight(TreeNode node) {
|
||
// 结点高度等于最高子树高度 + 1
|
||
node.height = Math.max(height(node.left), height(node.right)) + 1;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
### 结点平衡因子
|
||
|
||
结点的「平衡因子 Balance Factor」是 **结点的左子树高度减去右子树高度**,并定义空结点的平衡因子为 0 。同样地,我们将获取结点平衡因子封装成函数,以便后续使用。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 获取结点平衡因子 */
|
||
public int balanceFactor(TreeNode node) {
|
||
// 空结点平衡因子为 0
|
||
if (node == null) return 0;
|
||
// 结点平衡因子 = 左子树高度 - 右子树高度
|
||
return height(node.left) - height(node.right);
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
!!! note
|
||
|
||
设平衡因子为 $f$ ,则一棵 AVL 树的任意结点的平衡因子皆满足 $-1 \le f \le 1$ 。
|
||
|
||
## AVL 树旋转
|
||
|
||
AVL 树的独特之处在于「旋转 Rotation」的操作,其可 **在不影响二叉树中序遍历序列的前提下,使失衡结点重新恢复平衡。** 换言之,旋转操作既可以使树保持为「二叉搜索树」,也可以使树重新恢复为「平衡二叉树」。
|
||
|
||
我们将平衡因子的绝对值 $> 1$ 的结点称为「失衡结点」。根据结点的失衡情况,旋转操作分为 **右旋、左旋、先右旋后左旋、先左旋后右旋**,接下来我们来一起来看看它们是如何操作的。
|
||
|
||
### Case 1 - 右旋
|
||
|
||
如下图所示(结点下方为「平衡因子」),从底至顶看,二叉树中首个失衡结点是 **结点 2** 。我们聚焦在以结点 2 为根结点的子树上,将该结点记为 `node` ,将其左子节点记为 `child` ,执行「右旋」操作。完成右旋后,该子树已经恢复平衡,并且仍然为二叉搜索树。
|
||
|
||
=== "Step 1"
|
||
![right_rotate_step1](avl_tree.assets/right_rotate_step1.png)
|
||
=== "Step 2"
|
||
![right_rotate_step2](avl_tree.assets/right_rotate_step2.png)
|
||
=== "Step 3"
|
||
![right_rotate_step3](avl_tree.assets/right_rotate_step3.png)
|
||
=== "Step 4"
|
||
![right_rotate_step4](avl_tree.assets/right_rotate_step4.png)
|
||
|
||
进而,如果结点 `child` 本身有右子结点(记为 `grandChild`),则需要在「右旋」中添加一步:将 `grandChild` 作为 `node` 的左子结点。
|
||
|
||
![right_rotate_with_grandchild](avl_tree.assets/right_rotate_with_grandchild.png)
|
||
|
||
“向右旋转” 是一种形象化的说法,实际需要通过修改结点指针实现,代码如下所示。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 右旋操作 */
|
||
TreeNode rightRotate(TreeNode node) {
|
||
TreeNode child = node.left;
|
||
TreeNode grandChild = child.right;
|
||
// 以 child 为原点,将 node 向右旋转
|
||
child.right = node;
|
||
node.left = grandChild;
|
||
// 更新结点高度
|
||
updateHeight(node);
|
||
updateHeight(child);
|
||
// 返回旋转后的根节点
|
||
return child;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
### Case 2 - 左旋
|
||
|
||
类似地,如果将取上述失衡二叉树的 “镜像” ,那么则需要「左旋」操作。观察发现,**「左旋」和「右旋」操作是镜像对称的,两者对应解决的两种失衡情况也是对称的**,这说明两种旋转操作本质上是一样的。
|
||
|
||
![left_rotate_with_grandchild](avl_tree.assets/left_rotate_with_grandchild.png)
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 左旋操作 */
|
||
private TreeNode leftRotate(TreeNode node) {
|
||
TreeNode child = node.right;
|
||
TreeNode grandChild = child.left;
|
||
// 以 child 为原点,将 node 向左旋转
|
||
child.left = node;
|
||
node.right = grandChild;
|
||
// 更新结点高度
|
||
updateHeight(node);
|
||
updateHeight(child);
|
||
// 返回旋转后的根节点
|
||
return child;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
### Case 3 - 先左后右
|
||
|
||
对于下图的失衡结点 3 ,**单一使用左旋或右旋都无法使子树恢复平衡**,此时需要「先左旋后右旋」,即先对 `child` 执行「左旋」,再对 `node` 执行「右旋」。
|
||
|
||
![left_right_rotate](avl_tree.assets/left_right_rotate.png)
|
||
|
||
### Case 4 - 先右后左
|
||
|
||
同理,取以上失衡二叉树的镜像,则需要「先右旋后左旋」,即先对 `child` 执行「右旋」,然后对 `node` 执行「左旋」。
|
||
|
||
![right_left_rotate](avl_tree.assets/right_left_rotate.png)
|
||
|
||
### 旋转的选择
|
||
|
||
下图总结了以上四种失衡情况,分别采用右旋、左旋、先右后左、先左后右的旋转组合。
|
||
|
||
![rotation_cases](avl_tree.assets/rotation_cases.png)
|
||
|
||
具体地,需要使用 **失衡结点的平衡因子、较高一侧子结点的平衡因子** 来确定失衡结点属于上图中的哪种情况。
|
||
|
||
<div class="center-table" markdown>
|
||
|
||
| 失衡结点的平衡因子 | 子结点的平衡因子 | 应采用的旋转方法 |
|
||
| ------------------ | ---------------- | ---------------- |
|
||
| $>0$ (即左偏树) | $\geq 0$ | 右旋 |
|
||
| $>0$ (即左偏树) | $<0$ | 先左旋后右旋 |
|
||
| $<0$ (即右偏树) | $\leq 0$ | 左旋 |
|
||
| $<0$ (即右偏树) | $>0$ | 先右旋后左旋 |
|
||
|
||
</div>
|
||
|
||
下面,将旋转操作封装成一个函数。至此,**我们可以通过此函数来处理所有类型的失衡结点,使之恢复平衡**。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 执行旋转操作,使该子树重新恢复平衡 */
|
||
TreeNode rotate(TreeNode node) {
|
||
// 获取结点 node 的平衡因子
|
||
int balanceFactor = balanceFactor(node);
|
||
// 左偏树
|
||
if (balanceFactor > 1) {
|
||
if (balanceFactor(node.left) >= 0) {
|
||
// 右旋
|
||
return rightRotate(node);
|
||
} else {
|
||
// 先左旋后右旋
|
||
node.left = leftRotate(node.left);
|
||
return rightRotate(node);
|
||
}
|
||
}
|
||
// 右偏树
|
||
if (balanceFactor < -1) {
|
||
if (balanceFactor(node.right) <= 0) {
|
||
// 左旋
|
||
return leftRotate(node);
|
||
} else {
|
||
// 先右旋后左旋
|
||
node.right = rightRotate(node.right);
|
||
return leftRotate(node);
|
||
}
|
||
}
|
||
// 平衡树,无需旋转,直接返回
|
||
return node;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
## AVL 树常用操作
|
||
|
||
### 插入结点
|
||
|
||
「AVL 树」的结点插入操作与「二叉搜索树」主体类似。不同的是,在插入结点后,从该结点到根结点的路径上会出现「失衡结点」。所以,**我们需要从该结点开始,从底至顶地执行旋转操作,使所有失衡结点恢复平衡**。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 插入结点 */
|
||
TreeNode insert(int val) {
|
||
root = insertHelper(root, val);
|
||
return root;
|
||
}
|
||
|
||
/* 递归插入结点(辅助函数) */
|
||
TreeNode insertHelper(TreeNode node, int val) {
|
||
if (node == null) return new TreeNode(val);
|
||
/* 1. 查找插入位置,并插入结点 */
|
||
if (val < node.val)
|
||
node.left = insertHelper(node.left, val);
|
||
else if (val > node.val)
|
||
node.right = insertHelper(node.right, val);
|
||
else
|
||
return node; // 重复结点不插入,直接返回
|
||
updateHeight(node); // 更新结点高度
|
||
/* 2. 执行旋转操作,使该子树重新恢复平衡 */
|
||
node = rotate(node);
|
||
// 返回子树的根节点
|
||
return node;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
### 删除结点
|
||
|
||
「AVL 树」删除结点操作与「二叉搜索树」删除结点操作总体相同。类似地,**在删除结点后,也需要从底至顶地执行旋转操作,使所有失衡结点恢复平衡**。
|
||
|
||
=== "Java"
|
||
|
||
```java title="avl_tree.java"
|
||
/* 删除结点 */
|
||
TreeNode remove(int val) {
|
||
root = removeHelper(root, val);
|
||
return root;
|
||
}
|
||
|
||
/* 递归删除结点(辅助函数) */
|
||
TreeNode removeHelper(TreeNode node, int val) {
|
||
if (node == null) return null;
|
||
/* 1. 查找结点,并删除之 */
|
||
if (val < node.val)
|
||
node.left = removeHelper(node.left, val);
|
||
else if (val > node.val)
|
||
node.right = removeHelper(node.right, val);
|
||
else {
|
||
if (node.left == null || node.right == null) {
|
||
TreeNode child = node.left != null ? node.left : node.right;
|
||
// 子结点数量 = 0 ,直接删除 node 并返回
|
||
if (child == null)
|
||
return null;
|
||
// 子结点数量 = 1 ,直接删除 node
|
||
else
|
||
node = child;
|
||
} else {
|
||
// 子结点数量 = 2 ,则将中序遍历的下个结点删除,并用该结点替换当前结点
|
||
TreeNode temp = minNode(node.right);
|
||
node.right = removeHelper(node.right, temp.val);
|
||
node.val = temp.val;
|
||
}
|
||
}
|
||
updateHeight(node); // 更新结点高度
|
||
/* 2. 执行旋转操作,使该子树重新恢复平衡 */
|
||
node = rotate(node);
|
||
// 返回子树的根节点
|
||
return node;
|
||
}
|
||
|
||
/* 获取最小结点 */
|
||
TreeNode minNode(TreeNode node) {
|
||
if (node == null) return node;
|
||
// 循环访问左子结点,直到叶结点时为最小结点,跳出
|
||
while (node.left != null) {
|
||
node = node.left;
|
||
}
|
||
return node;
|
||
}
|
||
```
|
||
|
||
=== "C++"
|
||
|
||
```cpp title="avl_tree.cpp"
|
||
|
||
```
|
||
|
||
=== "Python"
|
||
|
||
```python title="avl_tree.py"
|
||
|
||
```
|
||
|
||
=== "Go"
|
||
|
||
```go title="avl_tree.go"
|
||
|
||
```
|
||
|
||
=== "JavaScript"
|
||
|
||
```js title="avl_tree.js"
|
||
|
||
```
|
||
|
||
=== "TypeScript"
|
||
|
||
```typescript title="avl_tree.ts"
|
||
|
||
```
|
||
|
||
=== "C"
|
||
|
||
```c title="avl_tree.c"
|
||
|
||
```
|
||
|
||
=== "C#"
|
||
|
||
```csharp title="avl_tree.cs"
|
||
|
||
```
|
||
|
||
### 查找结点
|
||
|
||
「AVL 树」的结点查找操作与「二叉搜索树」一致,在此不再赘述。
|