动态规划-PTA-0/1背包

Programming Contest
Bob will participate in a programming contest. There are altogether n problems in the contest. Unlike in PAT (Programming Ability Test), in a programming contest one can not obtain partial scores. For problem i, Bob will need time[i] to solve it and obtains the corresponding score[i], or he may choose not to solve it at all. Bob will be happy when he obtains a total score no less than happy_score. You are supposed to find the minimum time needed for Bob to be happy. The function need_time must return the minimum time, or -1 if it is impossible for Bob to obtain a score no less than happy_score.

Format of function:

int need_time(const int time[], const int score[], int happy_score, int n);

Here n (1≤n≤ MAXN) is the number of problems; happy_score (1≤ happy_score ≤ MAXS) is the minimum score for Bob to be happy; time[] is the array to store time[i] (1≤time[i]≤100) which is the time to solve problem i; score[] is the array to store score[i] (1≤score[i]≤100) which is the score Bob gets for solving problem i.

Sample program of judge:

#include <stdio.h>

#define MAXN 10
#define MAXS 1000

int need_time(const int time[], const int score[], int happy_score, int n);

int main() {
  int n, i, happy_score;
  int time[MAXN], score[MAXN];
  scanf("%d %d", &n, &happy_score);
  for (i = 0; i < n; ++i)
      scanf("%d", &time[i]);
  for (i = 0; i < n; ++i)
      scanf("%d", &score[i]);
  printf("%d\n", need_time(time, score, happy_score, n));
  return 0;
}

/* Your function will be put here */

Sample Input:

6 121

84 87 78 16 94 38
87 93 50 22 63 28

Sample Output:

125

本题是0/1背包问题的变种，首先介绍0/1背包问题：

0/1背包问题背景

给定背包的承重W，一堆物品的重量weights[ ]和价值values[ ]，求背包能装入物品的最大总value

0/1背包问题解法

定义一个二维数组 dp[物品总数][承重+1] 存储当前value，其中 dp[i][j] 表示：从前 i 件物品中找出一些放入背包，在总重不超过承重W的情况下，能达到的最大value。
第 i 件物品重量为 weight[i]，价值为 value[i]，根据第 i 件物品是否添加到背包中，可以分两种情况讨论：
① 第 i 件物品的重量超过背包承重，即 weight[i] > j。此时就算拿出背包中的所有物品也无法放下，所以它不可能被放入背包，因此有：dp[i][j] = dp[i-1][j]
② 第 i 件物品可以添加到背包中（可能拿出一些已有物品）。此时也可以有放入与不放入两种情况，应取value更大的那种。
（2.1）若不放入，有：dp[i][j] = dp[i-1][j]
（2.2）若放入，情况变为从某个状态下加入第i件物品并达到dp[i][j]结果，可知前一状态时的背包承重应为 j-weight[i]。因此有：dp[i][j] = dp[i-1][ j-weight[i] ] + value[i]。

综上，递推式为：

   dp[i][j] = max( dp[i-1][j], max( dp[i-1][ j-weight[i] ] + value[i], dp[i-1][j] ) );
即 dp[i][j] = max( dp[i-1][ j-weight[i] ] + value[i], dp[i-1][j] );

dp数组的一维简化：

可以看到不管第 i 件物品有没有加到背包中，新一轮 i 下的dp值都只由 i-1 决定，并且这种决定对于 j 是单方向的（即较大的 j 处的值只和较小或相同的 j 的值有关），因此若 j 从W -> 0 递减，dp的数组可以依次覆盖。

0/1背包问题的代码：

int* dp = (int*)malloc((W+1)*sizeof(int));
for (int i = 0; i <= W; i++) dp[i] = 0;

for (int i = 0; i <= n; i++) { 			// n = 物品数
	for (int j = W; j >= 1; j--) { 		// 逆序遍历
		if (j >= weights[i]) {
			dp[j] = max(dp[j], dp[ j-weights[i] ] + values[i] );
		}
	}
}
return dp[W];

回到原问题

本问题是求在 sum_score >= happy_score 情况下，sum_time 的最小值。
事实上:

• 做的题的score >= happy_score <=> 未做题的score < sum_score - happy_score
• 做的题的 (time) min <=> 未做题的 (time) max

所以原题等价于求出一些没做的题，使得，在这些题的总score不超过sum_score - happy_score情况下这些题的总time取得最大值。
所以：

• 未做题的总score即为物品的weights
• sum_score - happy_score即为背包的承重
• 未做题的总time即为物品的总value

这样我们完成了原问题与0/1背包问题的完美双射。

#include <stdlib.h>
#define max(x,y) ((x)>(y)?(x):(y))

int need_time(const int time[], const int score[], int happy_score, int n)
{
	int sum_time = 0;
	for (int i = 0; i < n; i++) sum_time += time[i];
	int sum_score = 0;
	for (int i = 0; i < n; i++) sum_score += score[i];
	
	if (sum_score < happy_score) return -1;

	int n_happy_score = sum_score - happy_score;

	int* dp = (int*)malloc((n_happy_score+1) * sizeof(int));
	for (int i = 0; i <= n_happy_score; i++) dp[i] = 0;

	for (int i = 0; i < n; i++) {
		for (int j = n_happy_score; j >= 0; j--) {
			if(j >= score[i])
				dp[j] = max(dp[j], dp[j - score[i]] + time[i]);
		}
	}

	return sum_time - dp[n_happy_score];
}