codeforces 183D

D. T-shirt
time limit per test5 seconds
memory limit per test256 megabytes
inputstandard input
outputstandard output
You are going to work in Codeforces as an intern in a team of n engineers, numbered 1 through n. You want to give each engineer a souvenir: a T-shirt from your country (T-shirts are highly desirable in Codeforces). Unfortunately you don’t know the size of the T-shirt each engineer fits in. There are m different sizes, numbered 1 through m, and each engineer will fit in a T-shirt of exactly one size.

You don’t know the engineers’ exact sizes, so you asked your friend, Gerald. Unfortunately, he wasn’t able to obtain the exact sizes either, but he managed to obtain for each engineer i and for all sizes j, the probability that the size of the T-shirt that fits engineer i is j.

Since you’re planning to give each engineer one T-shirt, you are going to bring with you exactly n T-shirts. For those n T-shirts, you can bring any combination of sizes (you can bring multiple T-shirts with the same size too!). You don’t know the sizes of T-shirts for each engineer when deciding what sizes to bring, so you have to pick this combination based only on the probabilities given by your friend, Gerald.

Your task is to maximize the expected number of engineers that receive a T-shirt of his size.

This is defined more formally as follows. When you finally arrive at the office, you will ask each engineer his T-shirt size. Then, if you still have a T-shirt of that size, you will give him one of them. Otherwise, you don’t give him a T-shirt. You will ask the engineers in order starting from engineer 1, then engineer 2, and so on until engineer n.

Input
The first line contains two space-separated integers n and m (1 ≤ n ≤ 3000, 1 ≤ m ≤ 300), denoting the number of engineers and the number of T-shirt sizes, respectively.

Then n lines follow, each line contains m space-separated integers. The j-th integer in the i-th line represents the probability that the i-th engineer fits in a T-shirt of size j. Each probability will be given as an integer between 0 and 1000, inclusive. The actual probability should be calculated as the given number divided by 1000.

It is guaranteed that for any engineer, the sum of the probabilities for all m T-shirts is equal to one.

Output
Print a single real number denoting the maximum possible expected number of engineers that will receive a T-shirt.

For the answer the absolute or relative error of 10 - 9 is acceptable.

题意描述：
$Description$
小A要给N个人准备礼物——T-shirt!但是你不知道他们的尺码……总共有M种尺码，编号从1到M。虽然你不记得每个人准确的尺码，但是你记得对于每一个人i，每一个尺码j，i的尺码正好是j的概率Pij.
现在你要买正好N件T-shirt，求能收到合适尺码礼物的人数的最大期望值。
你送礼物的方式是：从第1个人到第N个人依次询问，如果还有他们的尺码的T-shirt，就送出，否则就不送。
$Input$
第一行有两个整数N，M
接下来有N行，每行有M个整数，第i行的第j个整数表示Pij，用0到1000的整数表示，真实的概率用给出的数字除以1000得到。
保证每一行的整数和是1000。

首先我们可以发现每种礼物（暂且称T恤为礼物）都是独立的，互不影响的

我们考虑第$k$种礼物

令$f[i][j]$表示前$i$个人有$j$个人适合的概率
假设$p$为概率数组（已经转化为实数）

写出状态转移方程
$f[i][j] = f[i-1][j-1]*p[i][k] + f[i-1][j] * ( 1-p[i][k])$
但是再把k这一维加进去是不是MLE了啊？？
不急我们往下看

同样对于第$k$种礼物
$g[i]$表示买$i$件礼物能带来的期望
$g[i] = \sum^{i}_{j = 0}j * f[n][j] + i * \sum^{n}_{j = i + 1}f[n][j]$

这里如果我们不考虑满分，则可直接分组背包合并答案
时间复杂度$O(n^2m)$

这个做法的瓶颈就在于$f$数组的大小和合并

我们看$g$数组的计算形势
随着$增大$ ，相当于从右边取出一项移到右边。
我们不妨尝试对相邻两项做差，有惊喜！
$g[i+1]-g[i] = 1-\sum^{i}_{j = 1}f[n][j]$
我们可以发现这个插值是单调的
也就是$g$数组是上凸的！
所以$g[i]$不选$g[i+1]$也不选

我们考虑贪心
假设我们计算出了所有的$g[1]$我们显然取出$g$最大的那组物品最优
那我们取出来之后根据相邻两项的差算出下一项，减去前一项的贡献
再和原来的其它$g$比，就好了
这里我们可以发现$f$数组并不是全有用，我们只要随着选的物品增多同步更新$f$数组即可！
可以省掉一维和大量时间复杂度！
最终时间复杂度：$O（n^2 + nm）$
对于后面我们可以优化到$O(n^2+nlogm)$
没有必要这里就不过多篆述

代码如下：

#include<bits/stdc++.h>
using namespace std;

int n , m;
double p[3010][310] , f[310][3010];
double g[3010] , g_lazy[310] , tmp[3010]; 
double ans;
int num[310];
int read(){
    int sum = 0;char c = getchar();bool flag = true;
    while( c < '0' || c > '9' ) {if(c == '-') flag = false;c = getchar();}
    while( c >= '0' && c <= '9' ) sum = sum * 10 + c - 48 , c = getchar();
    if(flag)  return sum;
     else return -sum;
}  

void pre(){
    n = read();m = read();
    for(int i = 1;i <= n;++i)
        for(int j = 1;j <= m;++j)
            p[i][j] = read() * 1.0 / 1000;
    for(int j = 1;j <= m;++j)
        f[j][0] = 1.0;
    for(int i = 1;i <= m;++i){
        for(int j = 1;j <= n;++j)
            f[i][j] = f[i][j - 1] * ( 1 - p[j][i]);
        g_lazy[i] = 1.0 - f[i][n];
    }
    return;
}

void solve(int x){
    num[x]++;
    if(num[x] >= n){
        g_lazy[x] = 0;
        return;
    }
    for(int i = 0;i <= n;++i)
        tmp[i] = f[x][i];
    f[x][0] = 0.0;
    for(int i = 1;i <= n;++i)
        f[x][i] = f[x][i - 1] * (1.0 - p[i][x]) + tmp[i - 1] * p[i][x];
    g_lazy[x] -= f[x][n];
    return;
}

void work(){
    for(int i = 1;i <= n;++i){
        double Max = 0 ;int id = 0;
        for(int j = 1;j <= m;++j)
            if( Max < g_lazy[j] ){
                Max = g_lazy[j];
                id = j;
            }
        ans += Max;
        if(!id) return;
        solve(id);
    }
    printf("%.8lf",ans);
    return;
}

int main(){
    freopen("gift.in","r",stdin);
    freopen("gift.out","w",stdout); 
    pre();
    work();
    return 0;
}