完全背包问题

完全背包问题

---

动态规划

完全背包问题相对于01背包问题改变的就是第$i$个物品选取的数量，01背包最多就只能选一个，而完全背包只需要在体积允许的条件下选取任意个数，然后取$max$即为只从前$i$个物品中选且总体积不超过$j$的集合的最大值。
数组：$f[i][j]$
状态转移方程： f [ i ] [ j ] = m a x { f [ i ] [ j ] , f [ i − 1 ] [ j − k ∗ v [ i ] ] + k ∗ w [ i ] } f[i][j] = max\{f[i][j], f[i - 1][j - k * v[i]] + k * w[i]\} f[i][j]=max{f[i][j],f[i−1][j−k∗v[i]]+k∗w[i]} $(k从0开始枚举)$

$Java$ 代码

import java.util.*;

public class Main {
    static final int N = 1010;
    static int[] v = new int[N], w = new int[N];
    static int[][] f = new int[N][N];
    public static void main(String[] args) {
        Scanner sc = new Scanner(System.in);
        int n = sc.nextInt(), m = sc.nextInt();
        for (int i = 1; i <= n; i ++) {
            v[i] = sc.nextInt();
            w[i] = sc.nextInt();
        }
        for (int i = 1; i <= n; i ++) {
            for (int j = 1; j <= m; j ++) {
                for (int k = 0; k * v[i] <= j; k ++) {
                    f[i][j] = Math.max(f[i][j], f[i - 1][j - k * v[i]] + k * w[i]);
                }
            }
        }
        System.out.print(f[n][m]);
    }
}

---

优化时间

我们发现，上述的算法时间复杂度比较高，有3重循环，那么我们来考虑是否可以优化。

从$f[i][j]与f[i][j-v]$的关系出发，看能否推导除直接的关系，那么就可以省略$k$这一层的循环了。

f [ i ] [ j ] = m a x { f [ i − 1 ] [ j ] , f [ i − 1 ] [ j − v ] + w , f [ i − 1 ] [ j − 2 v ] + 2 w , . . . , f [ i − 1 ] [ j − k 1 v ] + k 1 w } f[i][j] = max\{f[i - 1][j], f[i - 1][j - v] + w, f[i - 1][j - 2v] + 2w, ..., f[i - 1][j - k_1v] + k_1w\} f[i][j]=max{f[i−1][j],f[i−1][j−v]+w,f[i−1][j−2v]+2w,...,f[i−1][j−k1​v]+k1​w}

f [ i ] [ j − v ] = m a x { f [ i − 1 ] [ j − v ] , f [ i − 1 ] [ j − 2 v ] + w , . . . , f [ i − 1 ] [ j − ( v + k 2 v ) ] + k 2 w } f[i][j - v] = max\{f[i - 1][j - v], f[i - 1][j - 2v] + w, ..., f[i - 1][j - (v + k_2v)] + k_2w\} f[i][j−v]=max{f[i−1][j−v],f[i−1][j−2v]+w,...,f[i−1][j−(v+k2​v)]+k2​w}

由$k_1,k_2$的含义可知，$k_{1}v=j$, $v+k_{2}v=j$, 所以化简得$k_1=k_2+1$。

故， f [ i ] [ j − v ] = m a x { f [ i − 1 ] [ j − v ] , f [ i − 1 ] [ j − 2 v ] + w , . . . , f [ i − 1 ] [ j − ( v + ( k 1 − 1 ) v ) ] + ( k 1 − 1 ) w } f[i][j - v] = max\{f[i - 1][j - v], f[i - 1][j - 2v] + w, ..., f[i - 1][j - (v + (k_1 - 1)v)] + (k_1 - 1)w\} f[i][j−v]=max{f[i−1][j−v],f[i−1][j−2v]+w,...,f[i−1][j−(v+(k1​−1)v)]+(k1​−1)w}

我们可以惊奇地发现，$f[i][j-v]$与$f[i][j]$除去第一项的其他部分，每一项之差为$w$，所以，

f [ i ] [ j ] = m a x { f [ i − 1 ] [ j ] , f [ i ] [ j − v ] + w } f[i][j] = max\{f[i - 1][j], f[i][j - v] + w\} f[i][j]=max{f[i−1][j],f[i][j−v]+w}

$Java$ 代码

import java.util.*;

public class Main {
    static final int N = 1010;
    static int[] v = new int[N], w = new int[N];
    static int[][] f = new int[N][N];
    public static void main(String[] args) {
        Scanner sc = new Scanner(System.in);
        int n = sc.nextInt(), m = sc.nextInt();
        for (int i = 1; i <= n; i ++) {
            v[i] = sc.nextInt();
            w[i] = sc.nextInt();
        }
        for (int i = 1; i <= n; i ++) 
            for (int j = 1; j <= m; j ++) {
                f[i][j] = f[i - 1][j];
                if (j >= v[i]) f[i][j] = Math.max(f[i - 1][j], f[i][j - v[i]] + w[i]);
            }
        System.out.print(f[n][m]);
    }
}

进一步优化空间

这里优化空间的思路同01背包，只是枚举$j$的顺序是从小到大，01背包是从大到小（需要理解原理）。

import java.util.*;

public class Main {
    static final int N = 1010;
    static int[] v = new int[N], w = new int[N], f = new int[N];
    public static void main(String[] args) {
        Scanner sc = new Scanner(System.in);
        int n = sc.nextInt(), m = sc.nextInt();
        for (int i = 1; i <= n; i ++) {
            v[i] = sc.nextInt();
            w[i] = sc.nextInt();
        }
        for (int i = 1; i <= n; i ++) 
            for (int j = v[i]; j <= m; j ++) {
                f[j] = Math.max(f[j], f[j - v[i]] + w[i]);
            }
        System.out.print(f[m]);
    }
}