mcpc2017 The Uncertainty of Politics

5096: The Uncertainty of Politics

时间限制: 1 Sec   内存限制: 128 MB   Special Judge

题目描述

You have an upcoming trip to Washington D.C. and you are fascinated with the intricacies of Congressional committee hearings. You wish to attend as many hearings as possible during your trip, and your local representative has provided you with a pass that will get you into the audience of any hearing. But there are some challenges in planning your schedule.
Specifically:
1. There are many committees, and thus many hearings, some of which take place at overlapping times.
2. While the committees are extremely punctual in terms of when to start a hearing, they are notoriously unpredictable in terms of how long the hearing lasts. Fortunately, rules do not actually allow for a filibuster of a committee hearing, so they cannot last forever.
3. It is considered rude to enter a hearing that is already underway, or to leave a hearing before it is complete. Given that you do not wish to embarrass the representative who provided your tickets, if you attend you must attend the entire hearing. Fortunately, hearings are very near to each other; as soon as one hearing is done, you can immediately join another hearing that is about to start.
Well in advance of your trip, Congress publishes a schedule of hearings, indicating for each one the time s at which the hearing will start, and then values a and b which represent, respectively, the shortest and longest possible length of that particular hearing. You are to assume that the actual length of the hearing will be a uniformly random integer over the inclusive interval [a, b]. 
Your goal is to develop a strategy that maximizes the expected number of hearings that you can attend during your trip. As an example, consider a situation in which there are four hearings with parameters as follows:

For this schedule, the optimal strategy will allow you to achieve an expected value of 2.125 hearings. To achieve this, you begin by attending the NASA hearing, which starts at time 3 and ends with equal probability at either time 5 or time 6 (given the hearing length that is uniformly distributed over {2, 3}). If the NASA hearing does end at time 5 you will immediately head to the oil and gas exploration hearing, and there is a  1/4 chance that hearing will end at time 6, allowing you to make yet a third hearing (about hurricane recovery efforts). If the NASA hearing instead ends at time 6, you will go straight to the hurricane hearing. 
By this strategy you will attend 3 hearings 12.5% of the time and 2 hearings the other 87.5% of the time, and thus expected value of 2.125. Note that if you were to start by attending the social media and elections hearing, you might optimistically make four hearings. However, a careful analysis will demonstrate that if you attend the first hearing, your optimal expected value is only 2.10714.

输入

The input begins with an integer n that designates the total number of scheduled hearings (1 ≤ n ≤ 10⁴).
Following that are n lines, each containing three integers s, a, and b, respectively representing the start time, minimum length, and maximum length of a hearing, such that 1 ≤ s ≤ 10⁶ and 1 ≤ a ≤ b ≤ 10⁶. The hearings will be listed in nondecreasing order of their start times.

输出

Display the expected number of hearings of an optimal strategy. Your answer should have an absolute or relative error of at most 10⁻³.

样例输入

4
1 1 7
3 2 3
5 1 4
6 10 10

样例输出

2.125

题目大意：有n个听证会，给定每个会议开始的时间，最早的结束时间和最晚的时间，求最明智的方案的参加听证会数目的期望值。

蒟蒻的思路：因为参加每一个听证会后最优方案的期望值只会收到它之后的听证会的最优方案的期望值的影响，所以第一反应就是从后往前进行一个DP。对每个听证会，将它的进行时间分为长度为1的m段，每段取在该时刻最大的期望，将一个听证会的所有时刻的最优值加在一起即为该听证会的最优解（如一个听证会进行时间是3-5时刻，那么我们分别取3、4、5这三个时刻能取得最大值，再将这三个值加在一起，最后加上该听证会自身的1，即为该听证会的最优解）

#include <bits/stdc++.h>
struct hearings
{
    int s,a,b;
};
double dp[100005];                    //dp[i]为参加第i个听证会（不参加第i个之前的听证会）的最优方案的期望值
hearings z[100005];
int main()
{
    int i,n;
    scanf("%d",&n);
    for (i=0;i<n;i++)
        scanf("%d%d%d",&z[i].s,&z[i].a,&z[i].b);
    for (i=n-1; i>=0; i--)           //从最后一个听证会往前扫
    {
        int p1,p2,e,k=n-1;
        double maxx=0;
        dp[i]++;                    //选择第i个听证会则他本身期望为1
        e=z[i].s+z[i].a;            //最早结束时间
        p2=z[i].s+z[i].b;            //p2为该听证会进行时间我们还未处理的最晚时刻（初始化为最晚结束时间）
        int m=z[i].b-z[i].a+1;        //听证会进行时间分为m段（每段长度）
        while (p2>=e)                //如果该听证会我们尚未处理完
        {
            while(k>i)                //k为开始时间晚于我们未处理的时刻的最早听证会
            {
                if (z[k].s<p2)
                    break;
                if (dp[k]>maxx)        //更新这个时刻我们能取的最优解
                    maxx=dp[k];        
                k--;
            }
            p1=e>z[k].s+1?e:z[k].s+1;    //将最优解不需要更新的时刻合并处理（不处理会TLE）（因为如果这一段时刻没有开始的听证会，则最优解不变）
            dp[i]+=maxx*(p2-p1+1)/m;    
            p2=p1-1;                    //更新p2
        }
    }
    double maxx=0;
    for (i=0;i<n;i++)
        if (dp[i]>maxx)
            maxx=dp[i];
    printf("%.16f",maxx);
    return 0;
}