The UNAL programming coaches have lost a bet, they bet the 6 UNAL teams would occupy the first six positions in the regional finals. Now, they have to shave their heads!. Some of them have more hair than others, and consequently, it takes more time to shave a head completely. However, all of the coaches really love their hair, therefore there is a probability that some (posibly all) of them daunt at the very last moment and do not permit the hairdresser to shave their heads.

Norbert, the hairdresser who loves probability, would like to order the coaches' schedule such that the average time in the hair salon is minimized for all the coaches. First all the coaches are there at the same time, then they start going one by one to Norbert, if by the moment a coach has to go to the hairdresser, he or she daunts then he or she simply leaves the hair salon and it is the turn of the next coach, after the head of a coach is shaved then that coach leaves the hair salon. The time between turns is negligible.

For example, suppose that shaving Diego's head takes 2 minutes and shaving Ivan's takes 3 minutes, but Diego has probability of 0.5 of not daunting meanwhile Ivan for sure will shave his head. If Ivan goes first, he will stay 3 minutes in the hair salon and Diego will stay there 3 minutes if daunting or 5 minutes if not (3 of them waiting for Ivan to finish), in this case the average expected time of the coaches in the hair salon would be 3.5, note this is not the optimal arrangement.

Now, Norbert knows the time it takes to shave each head and the probability of a coach to accept to have the head shaved in the barbershop, help him to know the minimum average expected time in the hair salon of the coaches.

Input

The first line of input is an integer n (1 ≤ n ≤ 5 * 105) - the number of coaches.

The next n lines contain each an integer x (0 ≤ x ≤ 100) and a decimal number y (0 ≤ y ≤ 1) separated by a single space - the time in minutes it takes to shave the head of the i - th coach and his probability of not daunting, respectively.

Output

Print the minimum expected average time. The relative error of your answer should not exceed 10 - 6.

Examples

Input
2
2 0.5
3 1.0
Output
2.500000000
Input
2
0 0.4
20 0.6
Output
6.000000000
从小到大排序,然后累计即可;
#include<bits/stdc++.h>
using namespace std;
#define maxn 600005
#define inf 999999999999
//#define INF 1e18
#define rdint(x) scanf("%d",&x)
#define rdllt(x) scanf("%lld",&x)
#define rdult(x) scanf("%lu",&x)
#define rdlf(x) scanf("%lf",&x)
#define rdstr(x) scanf("%s",x)
typedef long long ll;
const long long int mod=1e9+7;
#define ms(x) memset((x),0,sizeof(x)) int n;
struct node{
int x;
double prop;
}t[maxn];
double ans=0.0;
double tmp[maxn]; main(){
// ios_base::sync_with_stdio(0);cin.tie(0);cout.tie(0);
cin>>n;
for(int i=1;i<=n;i++)rdint(t[i].x),rdlf(t[i].prop),tmp[i]=1.0*t[i].x*t[i].prop;
sort(tmp+1,tmp+1+n);
for(int i=1;i<=n;i++){
ans+=tmp[i]*1.0*(n-i+1);
}
printf("%.7lf\n",1.0*ans/(double)n); }

Gym - 101845K 排序+概率的更多相关文章

  1. 【BZOJ3036】绿豆蛙的归宿 拓补排序+概率

    [BZOJ3036]绿豆蛙的归宿 Description 随着新版百度空间的下线,Blog宠物绿豆蛙完成了它的使命,去寻找它新的归宿. 给出一个有向无环的连通图,起点为1终点为N,每条边都有一个长度. ...

  2. 「10.28」Dove 打扑克(链表)·Cicada 与排序(概率)·Cicada 拿衣服(各种数据结构)

    A. Dove 打扑克 考场思考半天线段树树状数组,没有什么想法 打完暴力后突然想到此题用链表实现会很快. 因为只有$n$堆,所以设最多有$x$个不同的堆数,那么$x\times (x-1)/2==n ...

  3. A - Arcade Game Gym - 100814A (概率思维题)

    题目链接:https://cn.vjudge.net/contest/285964#problem/A 题目大意:每一次给你你一个数,然后对于每一次操作,可以将当前的数的每一位互换,如果互换后的数小于 ...

  4. Codeforces Round #253 (Div. 1) (A, B, C)

    Codeforces Round #253 (Div. 1) 题目链接 A:给定一些牌,然后如今要提示一些牌的信息,要求提示最少,使得全部牌能够被分辨出来. 思路:一共2^10种情况,直接暴力枚举,然 ...

  5. Learning To Rank之LambdaMART前世今生

    1.       前言 我们知道排序在非常多应用场景中属于一个非常核心的模块.最直接的应用就是搜索引擎.当用户提交一个query.搜索引擎会召回非常多文档,然后依据文档与query以及用户的相关程度对 ...

  6. [笔记]Learning to Rank算法介绍:RankNet,LambdaRank,LambdaMart

    之前的博客:http://www.cnblogs.com/bentuwuying/p/6681943.html中简单介绍了Learning to Rank的基本原理,也讲到了Learning to R ...

  7. CVPR2018: Unsupervised Cross-dataset Person Re-identification by Transfer Learning of Spatio-temporal Patterns

    论文可以在arxiv下载,老板一作,本人二作,也是我们实验室第一篇CCF A类论文,这个方法我们称为TFusion. 代码:https://github.com/ahangchen/TFusion 解 ...

  8. Learning to Rank算法介绍:RankNet,LambdaRank,LambdaMart

    之前的博客:http://www.cnblogs.com/bentuwuying/p/6681943.html中简单介绍了Learning to Rank的基本原理,也讲到了Learning to R ...

  9. 2016"百度之星" - 初赛(Astar Round2A)Gym Class(拓扑排序)

    Gym Class  Accepts: 849  Submissions: 4247  Time Limit: 6000/1000 MS (Java/Others)  Memory Limit: 65 ...

随机推荐

  1. svn使用技巧一:更新、提交、资源库同步之间区别

    提交:是用本地文件覆盖服务器的文件,只有提交会导致服务器上发生变化 更新:只是把服务器上最新版本下载到客户端,规则如下: 1.如果你本地的某个文件没有修改过,而服务器上的这个文件别人已经提交过新版本, ...

  2. Python打造一个目录扫描工具

    目标:用Python3写一款小型的web目录扫描工具 功能:1.扫描指定站点 2.指定网站脚本类型来扫描 3.可控线程 4.可保存扫描结果 首先定义一个命令参数的函数 def parse_option ...

  3. PowerDesigner中CDM和PDM如何定义外键关系

    有A.B两张表(实体),各自有id作为主键,两表是一一对应关系.但略有不同: A表一条记录可以对应0或1条B表记录,B表一条记录必须对应唯一条A表记录. 这样的关系如何在CDM或PDM中定义? 在最后 ...

  4. ansible for devops读书笔记第一章

    yum -y install ansible ansible --version mkdir /etc/ansible touch /etc/ansible/hosts [example]   www ...

  5. ks8基础(1) etcd安装

    下载安装 https://github.com/coreos/etcd/releases 在这网页,可以看到有多个版本共选择. 下载3.25 解压后, cd etcd-v3.2.5-linux-amd ...

  6. LAMP 3.4 mysql常用操作-2

    给用户授权 > grant all on discuz.* to 'user1'@'192.168.1.%' identified by 'wangshaojun'; 指定库,用户名user1 ...

  7. tomcat使用manager管理app时需要身份验证问题

    我们可以通过图形用户界面来管理tomcat,启动tomcat,在地址栏中输入: Java代码 http://localhost:8080 就可以看见tomcat的欢迎页面,点击左边的tomcat ma ...

  8. java连接字符串操作,可用来向sql传值

    private static String concat(String tag,String Time) { // TODO Auto-generated method stub // return ...

  9. gitlab 添加ssh秘钥

    在创建新的ssh秘钥对之前,要先确认一下电脑中是否已经有了一对秘钥: Git Bash on Windows / GNU/Linux / macOS / PowerShell: cat ~/.ssh/ ...

  10. Linux 控制台/终端/tty/shell

    一.简介 使用linux已经有一段时间,却一直弄不明白这几个概念之间的区别.这些概念本身有着非常浓厚的历史气息,随着时代的发展,他们的含义也在发生改变,它们有些已经失去了最初的含义,但是它们的名字却被 ...