|
|
---
|
|
|
comments: true
|
|
|
---
|
|
|
|
|
|
# 14.2 动态规划问题特性
|
|
|
|
|
|
在上一节中,我们学习了动态规划是如何通过子问题分解来求解原问题的。实际上,子问题分解是一种通用的算法思路,在分治、动态规划、回溯中的侧重点不同。
|
|
|
|
|
|
- 分治算法递归地将原问题划分为多个相互独立的子问题,直至最小子问题,并在回溯中合并子问题的解,最终得到原问题的解。
|
|
|
- 动态规划也对问题进行递归分解,但与分治算法的主要区别是,动态规划中的子问题是相互依赖的,在分解过程中会出现许多重叠子问题。
|
|
|
- 回溯算法在尝试和回退中穷举所有可能的解,并通过剪枝避免不必要的搜索分支。原问题的解由一系列决策步骤构成,我们可以将每个决策步骤之前的子序列看作一个子问题。
|
|
|
|
|
|
实际上,动态规划常用来求解最优化问题,它们不仅包含重叠子问题,还具有另外两大特性:最优子结构、无后效性。
|
|
|
|
|
|
## 14.2.1 最优子结构
|
|
|
|
|
|
我们对爬楼梯问题稍作改动,使之更加适合展示最优子结构概念。
|
|
|
|
|
|
!!! question "爬楼梯最小代价"
|
|
|
|
|
|
给定一个楼梯,你每步可以上 $1$ 阶或者 $2$ 阶,每一阶楼梯上都贴有一个非负整数,表示你在该台阶所需要付出的代价。给定一个非负整数数组 $cost$ ,其中 $cost[i]$ 表示在第 $i$ 个台阶需要付出的代价,$cost[0]$ 为地面(起始点)。请计算最少需要付出多少代价才能到达顶部?
|
|
|
|
|
|
如图 14-6 所示,若第 $1$、$2$、$3$ 阶的代价分别为 $1$、$10$、$1$ ,则从地面爬到第 $3$ 阶的最小代价为 $2$ 。
|
|
|
|
|
|
![爬到第 3 阶的最小代价](dp_problem_features.assets/min_cost_cs_example.png){ class="animation-figure" }
|
|
|
|
|
|
<p align="center"> 图 14-6 爬到第 3 阶的最小代价 </p>
|
|
|
|
|
|
设 $dp[i]$ 为爬到第 $i$ 阶累计付出的代价,由于第 $i$ 阶只可能从 $i - 1$ 阶或 $i - 2$ 阶走来,因此 $dp[i]$ 只可能等于 $dp[i - 1] + cost[i]$ 或 $dp[i - 2] + cost[i]$ 。为了尽可能减少代价,我们应该选择两者中较小的那一个:
|
|
|
|
|
|
$$
|
|
|
dp[i] = \min(dp[i-1], dp[i-2]) + cost[i]
|
|
|
$$
|
|
|
|
|
|
这便可以引出最优子结构的含义:**原问题的最优解是从子问题的最优解构建得来的**。
|
|
|
|
|
|
本题显然具有最优子结构:我们从两个子问题最优解 $dp[i-1]$ 和 $dp[i-2]$ 中挑选出较优的那一个,并用它构建出原问题 $dp[i]$ 的最优解。
|
|
|
|
|
|
那么,上一节的爬楼梯题目有没有最优子结构呢?它的目标是求解方案数量,看似是一个计数问题,但如果换一种问法:“求解最大方案数量”。我们意外地发现,**虽然题目修改前后是等价的,但最优子结构浮现出来了**:第 $n$ 阶最大方案数量等于第 $n-1$ 阶和第 $n-2$ 阶最大方案数量之和。所以说,最优子结构的解释方式比较灵活,在不同问题中会有不同的含义。
|
|
|
|
|
|
根据状态转移方程,以及初始状态 $dp[1] = cost[1]$ 和 $dp[2] = cost[2]$ ,我们就可以得到动态规划代码:
|
|
|
|
|
|
=== "Python"
|
|
|
|
|
|
```python title="min_cost_climbing_stairs_dp.py"
|
|
|
def min_cost_climbing_stairs_dp(cost: list[int]) -> int:
|
|
|
"""爬楼梯最小代价:动态规划"""
|
|
|
n = len(cost) - 1
|
|
|
if n == 1 or n == 2:
|
|
|
return cost[n]
|
|
|
# 初始化 dp 表,用于存储子问题的解
|
|
|
dp = [0] * (n + 1)
|
|
|
# 初始状态:预设最小子问题的解
|
|
|
dp[1], dp[2] = cost[1], cost[2]
|
|
|
# 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in range(3, n + 1):
|
|
|
dp[i] = min(dp[i - 1], dp[i - 2]) + cost[i]
|
|
|
return dp[n]
|
|
|
```
|
|
|
|
|
|
=== "C++"
|
|
|
|
|
|
```cpp title="min_cost_climbing_stairs_dp.cpp"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
int minCostClimbingStairsDP(vector<int> &cost) {
|
|
|
int n = cost.size() - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
vector<int> dp(n + 1);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i] = min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Java"
|
|
|
|
|
|
```java title="min_cost_climbing_stairs_dp.java"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
int minCostClimbingStairsDP(int[] cost) {
|
|
|
int n = cost.length - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int[] dp = new int[n + 1];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i] = Math.min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C#"
|
|
|
|
|
|
```csharp title="min_cost_climbing_stairs_dp.cs"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
int MinCostClimbingStairsDP(int[] cost) {
|
|
|
int n = cost.Length - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int[] dp = new int[n + 1];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i] = Math.Min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Go"
|
|
|
|
|
|
```go title="min_cost_climbing_stairs_dp.go"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
func minCostClimbingStairsDP(cost []int) int {
|
|
|
n := len(cost) - 1
|
|
|
if n == 1 || n == 2 {
|
|
|
return cost[n]
|
|
|
}
|
|
|
min := func(a, b int) int {
|
|
|
if a < b {
|
|
|
return a
|
|
|
}
|
|
|
return b
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
dp := make([]int, n+1)
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1]
|
|
|
dp[2] = cost[2]
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i := 3; i <= n; i++ {
|
|
|
dp[i] = min(dp[i-1], dp[i-2]) + cost[i]
|
|
|
}
|
|
|
return dp[n]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Swift"
|
|
|
|
|
|
```swift title="min_cost_climbing_stairs_dp.swift"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
func minCostClimbingStairsDP(cost: [Int]) -> Int {
|
|
|
let n = cost.count - 1
|
|
|
if n == 1 || n == 2 {
|
|
|
return cost[n]
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
var dp = Array(repeating: 0, count: n + 1)
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1]
|
|
|
dp[2] = cost[2]
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in stride(from: 3, through: n, by: 1) {
|
|
|
dp[i] = min(dp[i - 1], dp[i - 2]) + cost[i]
|
|
|
}
|
|
|
return dp[n]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "JS"
|
|
|
|
|
|
```javascript title="min_cost_climbing_stairs_dp.js"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
function minCostClimbingStairsDP(cost) {
|
|
|
const n = cost.length - 1;
|
|
|
if (n === 1 || n === 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
const dp = new Array(n + 1);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
dp[i] = Math.min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "TS"
|
|
|
|
|
|
```typescript title="min_cost_climbing_stairs_dp.ts"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
function minCostClimbingStairsDP(cost: Array<number>): number {
|
|
|
const n = cost.length - 1;
|
|
|
if (n === 1 || n === 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
const dp = new Array(n + 1);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
dp[i] = Math.min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Dart"
|
|
|
|
|
|
```dart title="min_cost_climbing_stairs_dp.dart"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
int minCostClimbingStairsDP(List<int> cost) {
|
|
|
int n = cost.length - 1;
|
|
|
if (n == 1 || n == 2) return cost[n];
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
List<int> dp = List.filled(n + 1, 0);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i] = min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Rust"
|
|
|
|
|
|
```rust title="min_cost_climbing_stairs_dp.rs"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
fn min_cost_climbing_stairs_dp(cost: &[i32]) -> i32 {
|
|
|
let n = cost.len() - 1;
|
|
|
if n == 1 || n == 2 { return cost[n]; }
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
let mut dp = vec![-1; n + 1];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in 3..=n {
|
|
|
dp[i] = cmp::min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
dp[n]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C"
|
|
|
|
|
|
```c title="min_cost_climbing_stairs_dp.c"
|
|
|
/* 爬楼梯最小代价:动态规划 */
|
|
|
int minCostClimbingStairsDP(int cost[], int costSize) {
|
|
|
int n = costSize - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int *dp = calloc(n + 1, sizeof(int));
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i] = myMin(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
int res = dp[n];
|
|
|
// 释放内存
|
|
|
free(dp);
|
|
|
return res;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Zig"
|
|
|
|
|
|
```zig title="min_cost_climbing_stairs_dp.zig"
|
|
|
// 爬楼梯最小代价:动态规划
|
|
|
fn minCostClimbingStairsDP(comptime cost: []i32) i32 {
|
|
|
comptime var n = cost.len - 1;
|
|
|
if (n == 1 or n == 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
var dp = [_]i32{-1} ** (n + 1);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1] = cost[1];
|
|
|
dp[2] = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (3..n + 1) |i| {
|
|
|
dp[i] = @min(dp[i - 1], dp[i - 2]) + cost[i];
|
|
|
}
|
|
|
return dp[n];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
图 14-7 展示了以上代码的动态规划过程。
|
|
|
|
|
|
![爬楼梯最小代价的动态规划过程](dp_problem_features.assets/min_cost_cs_dp.png){ class="animation-figure" }
|
|
|
|
|
|
<p align="center"> 图 14-7 爬楼梯最小代价的动态规划过程 </p>
|
|
|
|
|
|
本题也可以进行空间优化,将一维压缩至零维,使得空间复杂度从 $O(n)$ 降至 $O(1)$ :
|
|
|
|
|
|
=== "Python"
|
|
|
|
|
|
```python title="min_cost_climbing_stairs_dp.py"
|
|
|
def min_cost_climbing_stairs_dp_comp(cost: list[int]) -> int:
|
|
|
"""爬楼梯最小代价:空间优化后的动态规划"""
|
|
|
n = len(cost) - 1
|
|
|
if n == 1 or n == 2:
|
|
|
return cost[n]
|
|
|
a, b = cost[1], cost[2]
|
|
|
for i in range(3, n + 1):
|
|
|
a, b = b, min(a, b) + cost[i]
|
|
|
return b
|
|
|
```
|
|
|
|
|
|
=== "C++"
|
|
|
|
|
|
```cpp title="min_cost_climbing_stairs_dp.cpp"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
int minCostClimbingStairsDPComp(vector<int> &cost) {
|
|
|
int n = cost.size() - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
int a = cost[1], b = cost[2];
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
int tmp = b;
|
|
|
b = min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Java"
|
|
|
|
|
|
```java title="min_cost_climbing_stairs_dp.java"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
int minCostClimbingStairsDPComp(int[] cost) {
|
|
|
int n = cost.length - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
int a = cost[1], b = cost[2];
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
int tmp = b;
|
|
|
b = Math.min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C#"
|
|
|
|
|
|
```csharp title="min_cost_climbing_stairs_dp.cs"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
int MinCostClimbingStairsDPComp(int[] cost) {
|
|
|
int n = cost.Length - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
int a = cost[1], b = cost[2];
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
int tmp = b;
|
|
|
b = Math.Min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Go"
|
|
|
|
|
|
```go title="min_cost_climbing_stairs_dp.go"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
func minCostClimbingStairsDPComp(cost []int) int {
|
|
|
n := len(cost) - 1
|
|
|
if n == 1 || n == 2 {
|
|
|
return cost[n]
|
|
|
}
|
|
|
min := func(a, b int) int {
|
|
|
if a < b {
|
|
|
return a
|
|
|
}
|
|
|
return b
|
|
|
}
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
a, b := cost[1], cost[2]
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i := 3; i <= n; i++ {
|
|
|
tmp := b
|
|
|
b = min(a, tmp) + cost[i]
|
|
|
a = tmp
|
|
|
}
|
|
|
return b
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Swift"
|
|
|
|
|
|
```swift title="min_cost_climbing_stairs_dp.swift"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
func minCostClimbingStairsDPComp(cost: [Int]) -> Int {
|
|
|
let n = cost.count - 1
|
|
|
if n == 1 || n == 2 {
|
|
|
return cost[n]
|
|
|
}
|
|
|
var (a, b) = (cost[1], cost[2])
|
|
|
for i in stride(from: 3, through: n, by: 1) {
|
|
|
(a, b) = (b, min(a, b) + cost[i])
|
|
|
}
|
|
|
return b
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "JS"
|
|
|
|
|
|
```javascript title="min_cost_climbing_stairs_dp.js"
|
|
|
/* 爬楼梯最小代价:状态压缩后的动态规划 */
|
|
|
function minCostClimbingStairsDPComp(cost) {
|
|
|
const n = cost.length - 1;
|
|
|
if (n === 1 || n === 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
let a = cost[1],
|
|
|
b = cost[2];
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
const tmp = b;
|
|
|
b = Math.min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "TS"
|
|
|
|
|
|
```typescript title="min_cost_climbing_stairs_dp.ts"
|
|
|
/* 爬楼梯最小代价:状态压缩后的动态规划 */
|
|
|
function minCostClimbingStairsDPComp(cost: Array<number>): number {
|
|
|
const n = cost.length - 1;
|
|
|
if (n === 1 || n === 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
let a = cost[1],
|
|
|
b = cost[2];
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
const tmp = b;
|
|
|
b = Math.min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Dart"
|
|
|
|
|
|
```dart title="min_cost_climbing_stairs_dp.dart"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
int minCostClimbingStairsDPComp(List<int> cost) {
|
|
|
int n = cost.length - 1;
|
|
|
if (n == 1 || n == 2) return cost[n];
|
|
|
int a = cost[1], b = cost[2];
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
int tmp = b;
|
|
|
b = min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Rust"
|
|
|
|
|
|
```rust title="min_cost_climbing_stairs_dp.rs"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
fn min_cost_climbing_stairs_dp_comp(cost: &[i32]) -> i32 {
|
|
|
let n = cost.len() - 1;
|
|
|
if n == 1 || n == 2 { return cost[n] };
|
|
|
let (mut a, mut b) = (cost[1], cost[2]);
|
|
|
for i in 3..=n {
|
|
|
let tmp = b;
|
|
|
b = cmp::min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
b
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C"
|
|
|
|
|
|
```c title="min_cost_climbing_stairs_dp.c"
|
|
|
/* 爬楼梯最小代价:空间优化后的动态规划 */
|
|
|
int minCostClimbingStairsDPComp(int cost[], int costSize) {
|
|
|
int n = costSize - 1;
|
|
|
if (n == 1 || n == 2)
|
|
|
return cost[n];
|
|
|
int a = cost[1], b = cost[2];
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
int tmp = b;
|
|
|
b = myMin(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Zig"
|
|
|
|
|
|
```zig title="min_cost_climbing_stairs_dp.zig"
|
|
|
// 爬楼梯最小代价:空间优化后的动态规划
|
|
|
fn minCostClimbingStairsDPComp(cost: []i32) i32 {
|
|
|
var n = cost.len - 1;
|
|
|
if (n == 1 or n == 2) {
|
|
|
return cost[n];
|
|
|
}
|
|
|
var a = cost[1];
|
|
|
var b = cost[2];
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (3..n + 1) |i| {
|
|
|
var tmp = b;
|
|
|
b = @min(a, tmp) + cost[i];
|
|
|
a = tmp;
|
|
|
}
|
|
|
return b;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
## 14.2.2 无后效性
|
|
|
|
|
|
无后效性是动态规划能够有效解决问题的重要特性之一,其定义为:**给定一个确定的状态,它的未来发展只与当前状态有关,而与过去经历的所有状态无关**。
|
|
|
|
|
|
以爬楼梯问题为例,给定状态 $i$ ,它会发展出状态 $i+1$ 和状态 $i+2$ ,分别对应跳 $1$ 步和跳 $2$ 步。在做出这两种选择时,我们无须考虑状态 $i$ 之前的状态,它们对状态 $i$ 的未来没有影响。
|
|
|
|
|
|
然而,如果我们给爬楼梯问题添加一个约束,情况就不一样了。
|
|
|
|
|
|
!!! question "带约束爬楼梯"
|
|
|
|
|
|
给定一个共有 $n$ 阶的楼梯,你每步可以上 $1$ 阶或者 $2$ 阶,**但不能连续两轮跳 $1$ 阶**,请问有多少种方案可以爬到楼顶?
|
|
|
|
|
|
如图 14-8 所示,爬上第 $3$ 阶仅剩 $2$ 种可行方案,其中连续三次跳 $1$ 阶的方案不满足约束条件,因此被舍弃。
|
|
|
|
|
|
![带约束爬到第 3 阶的方案数量](dp_problem_features.assets/climbing_stairs_constraint_example.png){ class="animation-figure" }
|
|
|
|
|
|
<p align="center"> 图 14-8 带约束爬到第 3 阶的方案数量 </p>
|
|
|
|
|
|
在该问题中,如果上一轮是跳 $1$ 阶上来的,那么下一轮就必须跳 $2$ 阶。这意味着,**下一步选择不能由当前状态(当前所在楼梯阶数)独立决定,还和前一个状态(上一轮所在楼梯阶数)有关**。
|
|
|
|
|
|
不难发现,此问题已不满足无后效性,状态转移方程 $dp[i] = dp[i-1] + dp[i-2]$ 也失效了,因为 $dp[i-1]$ 代表本轮跳 $1$ 阶,但其中包含了许多“上一轮是跳 $1$ 阶上来的”方案,而为了满足约束,我们就不能将 $dp[i-1]$ 直接计入 $dp[i]$ 中。
|
|
|
|
|
|
为此,我们需要扩展状态定义:**状态 $[i, j]$ 表示处在第 $i$ 阶并且上一轮跳了 $j$ 阶**,其中 $j \in \{1, 2\}$ 。此状态定义有效地区分了上一轮跳了 $1$ 阶还是 $2$ 阶,我们可以据此判断当前状态是从何而来的。
|
|
|
|
|
|
- 当上一轮跳了 $1$ 阶时,上上一轮只能选择跳 $2$ 阶,即 $dp[i, 1]$ 只能从 $dp[i-1, 2]$ 转移过来。
|
|
|
- 当上一轮跳了 $2$ 阶时,上上一轮可选择跳 $1$ 阶或跳 $2$ 阶,即 $dp[i, 2]$ 可以从 $dp[i-2, 1]$ 或 $dp[i-2, 2]$ 转移过来。
|
|
|
|
|
|
如图 14-9 所示,在该定义下,$dp[i, j]$ 表示状态 $[i, j]$ 对应的方案数。此时状态转移方程为:
|
|
|
|
|
|
$$
|
|
|
\begin{cases}
|
|
|
dp[i, 1] = dp[i-1, 2] \\
|
|
|
dp[i, 2] = dp[i-2, 1] + dp[i-2, 2]
|
|
|
\end{cases}
|
|
|
$$
|
|
|
|
|
|
![考虑约束下的递推关系](dp_problem_features.assets/climbing_stairs_constraint_state_transfer.png){ class="animation-figure" }
|
|
|
|
|
|
<p align="center"> 图 14-9 考虑约束下的递推关系 </p>
|
|
|
|
|
|
最终,返回 $dp[n, 1] + dp[n, 2]$ 即可,两者之和代表爬到第 $n$ 阶的方案总数:
|
|
|
|
|
|
=== "Python"
|
|
|
|
|
|
```python title="climbing_stairs_constraint_dp.py"
|
|
|
def climbing_stairs_constraint_dp(n: int) -> int:
|
|
|
"""带约束爬楼梯:动态规划"""
|
|
|
if n == 1 or n == 2:
|
|
|
return 1
|
|
|
# 初始化 dp 表,用于存储子问题的解
|
|
|
dp = [[0] * 3 for _ in range(n + 1)]
|
|
|
# 初始状态:预设最小子问题的解
|
|
|
dp[1][1], dp[1][2] = 1, 0
|
|
|
dp[2][1], dp[2][2] = 0, 1
|
|
|
# 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in range(3, n + 1):
|
|
|
dp[i][1] = dp[i - 1][2]
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2]
|
|
|
return dp[n][1] + dp[n][2]
|
|
|
```
|
|
|
|
|
|
=== "C++"
|
|
|
|
|
|
```cpp title="climbing_stairs_constraint_dp.cpp"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
int climbingStairsConstraintDP(int n) {
|
|
|
if (n == 1 || n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
vector<vector<int>> dp(n + 1, vector<int>(3, 0));
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Java"
|
|
|
|
|
|
```java title="climbing_stairs_constraint_dp.java"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
int climbingStairsConstraintDP(int n) {
|
|
|
if (n == 1 || n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int[][] dp = new int[n + 1][3];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C#"
|
|
|
|
|
|
```csharp title="climbing_stairs_constraint_dp.cs"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
int ClimbingStairsConstraintDP(int n) {
|
|
|
if (n == 1 || n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int[,] dp = new int[n + 1, 3];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1, 1] = 1;
|
|
|
dp[1, 2] = 0;
|
|
|
dp[2, 1] = 0;
|
|
|
dp[2, 2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i, 1] = dp[i - 1, 2];
|
|
|
dp[i, 2] = dp[i - 2, 1] + dp[i - 2, 2];
|
|
|
}
|
|
|
return dp[n, 1] + dp[n, 2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Go"
|
|
|
|
|
|
```go title="climbing_stairs_constraint_dp.go"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
func climbingStairsConstraintDP(n int) int {
|
|
|
if n == 1 || n == 2 {
|
|
|
return 1
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
dp := make([][3]int, n+1)
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1
|
|
|
dp[1][2] = 0
|
|
|
dp[2][1] = 0
|
|
|
dp[2][2] = 1
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i := 3; i <= n; i++ {
|
|
|
dp[i][1] = dp[i-1][2]
|
|
|
dp[i][2] = dp[i-2][1] + dp[i-2][2]
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Swift"
|
|
|
|
|
|
```swift title="climbing_stairs_constraint_dp.swift"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
func climbingStairsConstraintDP(n: Int) -> Int {
|
|
|
if n == 1 || n == 2 {
|
|
|
return 1
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
var dp = Array(repeating: Array(repeating: 0, count: 3), count: n + 1)
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1
|
|
|
dp[1][2] = 0
|
|
|
dp[2][1] = 0
|
|
|
dp[2][2] = 1
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in stride(from: 3, through: n, by: 1) {
|
|
|
dp[i][1] = dp[i - 1][2]
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2]
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "JS"
|
|
|
|
|
|
```javascript title="climbing_stairs_constraint_dp.js"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
function climbingStairsConstraintDP(n) {
|
|
|
if (n === 1 || n === 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
const dp = Array.from(new Array(n + 1), () => new Array(3));
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "TS"
|
|
|
|
|
|
```typescript title="climbing_stairs_constraint_dp.ts"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
function climbingStairsConstraintDP(n: number): number {
|
|
|
if (n === 1 || n === 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
const dp = Array.from({ length: n + 1 }, () => new Array(3));
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (let i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Dart"
|
|
|
|
|
|
```dart title="climbing_stairs_constraint_dp.dart"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
int climbingStairsConstraintDP(int n) {
|
|
|
if (n == 1 || n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
List<List<int>> dp = List.generate(n + 1, (index) => List.filled(3, 0));
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Rust"
|
|
|
|
|
|
```rust title="climbing_stairs_constraint_dp.rs"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
fn climbing_stairs_constraint_dp(n: usize) -> i32 {
|
|
|
if n == 1 || n == 2 { return 1 };
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
let mut dp = vec![vec![-1; 3]; n + 1];
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for i in 3..=n {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
dp[n][1] + dp[n][2]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "C"
|
|
|
|
|
|
```c title="climbing_stairs_constraint_dp.c"
|
|
|
/* 带约束爬楼梯:动态规划 */
|
|
|
int climbingStairsConstraintDP(int n) {
|
|
|
if (n == 1 || n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
int **dp = malloc((n + 1) * sizeof(int *));
|
|
|
for (int i = 0; i <= n; i++) {
|
|
|
dp[i] = calloc(3, sizeof(int));
|
|
|
}
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (int i = 3; i <= n; i++) {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
int res = dp[n][1] + dp[n][2];
|
|
|
// 释放内存
|
|
|
for (int i = 0; i <= n; i++) {
|
|
|
free(dp[i]);
|
|
|
}
|
|
|
free(dp);
|
|
|
return res;
|
|
|
}
|
|
|
```
|
|
|
|
|
|
=== "Zig"
|
|
|
|
|
|
```zig title="climbing_stairs_constraint_dp.zig"
|
|
|
// 带约束爬楼梯:动态规划
|
|
|
fn climbingStairsConstraintDP(comptime n: usize) i32 {
|
|
|
if (n == 1 or n == 2) {
|
|
|
return 1;
|
|
|
}
|
|
|
// 初始化 dp 表,用于存储子问题的解
|
|
|
var dp = [_][3]i32{ [_]i32{ -1, -1, -1 } } ** (n + 1);
|
|
|
// 初始状态:预设最小子问题的解
|
|
|
dp[1][1] = 1;
|
|
|
dp[1][2] = 0;
|
|
|
dp[2][1] = 0;
|
|
|
dp[2][2] = 1;
|
|
|
// 状态转移:从较小子问题逐步求解较大子问题
|
|
|
for (3..n + 1) |i| {
|
|
|
dp[i][1] = dp[i - 1][2];
|
|
|
dp[i][2] = dp[i - 2][1] + dp[i - 2][2];
|
|
|
}
|
|
|
return dp[n][1] + dp[n][2];
|
|
|
}
|
|
|
```
|
|
|
|
|
|
在上面的案例中,由于仅需多考虑前面一个状态,因此我们仍然可以通过扩展状态定义,使得问题重新满足无后效性。然而,某些问题具有非常严重的“有后效性”。
|
|
|
|
|
|
!!! question "爬楼梯与障碍生成"
|
|
|
|
|
|
给定一个共有 $n$ 阶的楼梯,你每步可以上 $1$ 阶或者 $2$ 阶。**规定当爬到第 $i$ 阶时,系统自动会在第 $2i$ 阶上放上障碍物,之后所有轮都不允许跳到第 $2i$ 阶上**。例如,前两轮分别跳到了第 $2$、$3$ 阶上,则之后就不能跳到第 $4$、$6$ 阶上。请问有多少种方案可以爬到楼顶?
|
|
|
|
|
|
在这个问题中,下次跳跃依赖过去所有的状态,因为每一次跳跃都会在更高的阶梯上设置障碍,并影响未来的跳跃。对于这类问题,动态规划往往难以解决。
|
|
|
|
|
|
实际上,许多复杂的组合优化问题(例如旅行商问题)不满足无后效性。对于这类问题,我们通常会选择使用其他方法,例如启发式搜索、遗传算法、强化学习等,从而在有限时间内得到可用的局部最优解。
|