Hard Life
Time Limit: 8000MS   Memory Limit: 65536K
Total Submissions: 9012   Accepted: 2614
Case Time Limit: 2000MS   Special Judge

Description

John is a Chief Executive Officer at a privately owned medium size company. The owner of the company has decided to make his son Scott a manager in the company. John fears that the owner will ultimately give CEO position to Scott if he does well on his new manager position, so he decided to make Scott’s life as hard as possible by carefully selecting the team he is going to manage in the company.

John knows which pairs of his people work poorly in the same team. John introduced a hardness factor of a team — it is a number of pairs of people from this team who work poorly in the same team divided by the total number of people in the team. The larger is the hardness factor, the harder is this team to manage. John wants to find a group of people in the company that are hardest to manage and make it Scott’s team. Please, help him.

In the example on the picture the hardest team consists of people 1, 2, 4, and 5. Among 4 of them 5 pairs work poorly in the same team, thus hardness factor is equal to 54. If we add person number 3 to the team then hardness factor decreases to 65.

Input

The first line of the input file contains two integer numbers n and m (1 ≤ n ≤ 100, 0 ≤ m ≤ 1000). Here n is a total number of people in the company (people are numbered from 1 to n), and m is the number of pairs of people who work poorly in the same team. Next m lines describe those pairs with two integer numbers ai and bi (1 ≤ aibi ≤ nai ≠ bi) on a line. The order of people in a pair is arbitrary and no pair is listed twice.

Output

Write to the output file an integer number k (1 ≤ k ≤ n) — the number of people in the hardest team, followed by k lines listing people from this team in ascending order. If there are multiple teams with the same hardness factor then write any one.

Sample Input

sample input #1
5 6
1 5
5 4
4 2
2 5
1 2
3 1 sample input #2
4 0

Sample Output

sample output #1
4
1
2
4
5 sample output #2
1
1

Hint

Note, that in the last example any team has hardness factor of zero, and any non-empty list of people is a valid answer.

题目链接:POJ 3155

官方解说在胡伯涛Amber的最小割模型在信息学竞赛中的应用论文中最大密度子图是什么意思呢?就是选出一个子图,使得其子图的${边数 \over 顶点数}$最大,即${|E'| \over |V'|}$最大,后者是不是非常的熟悉,跟01规划非常像,就是使得该比值最大,那么按照01规划的思路,设这个比值为$g$,那么二分出一个$g$的最大值使得$|E'|-g*|V'|>0$,但是我怎么知道$E'$和$V'$怎么取啊?再看式子里面的$E'$和$V'$,可以发现若一条边$e_i$在边集$E'$中,那么$e_i$的两个端点$u_i$与$v_i$必定在顶点集$V'$中,也就是说$e_i$存在的必要条件是$u_i$,$v_i$两点的存在,然后再次观察这个式子,又可以发现可以把边看成新的虚拟点,点还是原来的点,虚拟点的点权为正数1,原来点的点权为负数-g,那么只要在这个新图里求一个最大权闭合图的最大权值就是这个$|E'|-g*|V'|$的最大值即这个式子的最优情况,然后根据其与0(或者叫eps…)的关系进行二分即可,码农题……

代码:

#include <stdio.h>
#include <iostream>
#include <algorithm>
#include <cstdlib>
#include <sstream>
#include <numeric>
#include <cstring>
#include <bitset>
#include <string>
#include <deque>
#include <stack>
#include <cmath>
#include <queue>
#include <set>
#include <map>
using namespace std;
#define INF 0x3f3f3f3f
#define LC(x) (x<<1)
#define RC(x) ((x<<1)+1)
#define MID(x,y) ((x+y)>>1)
#define CLR(arr,val) memset(arr,val,sizeof(arr))
#define FAST_IO ios::sync_with_stdio(false);cin.tie(0);
typedef pair<int, int> pii;
typedef long long LL;
const double PI = acos(-1.0);
const int N = 110;
const int M = 1010;
const double eps = 1e-8;
struct edge
{
int to, nxt;
double cap;
edge() {}
edge(int _to, int _nxt, double _cap): to(_to), nxt(_nxt), cap(_cap) {}
};
struct info
{
int u, v;
};
info e[M];
edge E[(N + 3 * M) << 1];
int head[N + M], tot;
int d[N + M]; void init()
{
CLR(head, -1);
tot = 0;
}
inline void add(int s, int t, double cap)
{
E[tot] = edge(t, head[s], cap);
head[s] = tot++;
E[tot] = edge(s, head[t], 0);
head[t] = tot++;
}
int bfs(int s, int t)
{
queue<int>Q;
CLR(d, -1);
d[s] = 0;
Q.push(s);
while (!Q.empty())
{
int u = Q.front();
Q.pop();
for (int i = head[u]; ~i; i = E[i].nxt)
{
int v = E[i].to;
if (d[v] == -1 && E[i].cap > 0)
{
d[v] = d[u] + 1;
if (v == t)
return 1;
Q.push(v);
}
}
}
return ~d[t];
}
double dfs(int s, int t, double f)
{
if (s == t || f == 0)
return f;
double ret = 0;
for (int i = head[s]; ~i; i = E[i].nxt)
{
int v = E[i].to;
if (d[v] == d[s] + 1 && E[i].cap > 0)
{
double df = dfs(v, t, min(f, E[i].cap));
if (df > 0)
{
E[i].cap -= df;
E[i ^ 1].cap += df;
ret += df;
f -= df;
if (f == 0)
break;
}
}
}
if (ret == 0)
d[s] = -1;
return ret;
}
double dinic(int s, int t)
{
double ret = 0;
while (bfs(s, t))
ret += dfs(s, t, 1e9);
return ret;
}
double Maxweight(int n, int m, double g)
{
init();
int S = 0, T = n + m + 1;
int i;
for (i = 1; i <= n; ++i) //负权点连到汇点T
add(i, T, g);
for (i = 1; i <= m; ++i) //m个正权点从S连出
{
add(S, n + i, 1.0);
add(n + i, e[i].u, 1e9); //原有边保留
add(n + i, e[i].v, 1e9);
}
return m * 1.0 - dinic(S, T); //正权点-最大流得到最大权
}
int main(void)
{
int n, m, i;
while (~scanf("%d%d", &n, &m))
{
for (i = 1; i <= m; ++i)
scanf("%d%d", &e[i].u, &e[i].v);
if (!m)
{
puts("1");
puts("1");
}
else
{
double L = 1.0 / n, R = m;
double ans = 1;
double dx = 1.0 / n / n;
while (fabs(R - L) >= dx)
{
double mid = (L + R) / 2.0;
if (fabs(Maxweight(n, m, mid)) >= eps)
{
L = mid;
ans = mid;
}
else
R = mid;
}
Maxweight(n, m, ans);
vector<int>pos;
for (i = 1; i <= n; ++i)
if (~d[i])
pos.push_back(i);
int sz = pos.size();
sort(pos.begin(), pos.end());
printf("%d\n", sz);
for (i = 0; i < sz; ++i)
printf("%d\n", pos[i]);
}
}
return 0;
}

改进算法:由论文可得$h(g)={{U*n-c[S, T]} \over 2}$,其中可以令$U=m$,然后新的建图方式为

改进后从600ms到200ms,效果还是不错的

代码:

#include <stdio.h>
#include <iostream>
#include <algorithm>
#include <cstdlib>
#include <sstream>
#include <numeric>
#include <cstring>
#include <bitset>
#include <string>
#include <deque>
#include <stack>
#include <cmath>
#include <queue>
#include <set>
#include <map>
using namespace std;
#define INF 0x3f3f3f3f
#define LC(x) (x<<1)
#define RC(x) ((x<<1)+1)
#define MID(x,y) ((x+y)>>1)
#define CLR(arr,val) memset(arr,val,sizeof(arr))
#define FAST_IO ios::sync_with_stdio(false);cin.tie(0);
typedef pair<int, int> pii;
typedef long long LL;
const double PI = acos(-1.0);
const int N = 110;
const int M = 1010;
const double eps = 1e-8;
struct edge
{
int to, nxt;
double cap;
edge() {}
edge(int _to, int _nxt, double _cap): to(_to), nxt(_nxt), cap(_cap) {}
};
struct info
{
int u, v;
};
info e[M];
edge E[(N + M) << 2];
int head[N], tot;
int d[N], deg[N]; void init()
{
CLR(head, -1);
tot = 0;
}
inline void add(int s, int t, double cap)
{
E[tot] = edge(t, head[s], cap);
head[s] = tot++;
E[tot] = edge(s, head[t], 0);
head[t] = tot++;
}
int bfs(int s, int t)
{
queue<int>Q;
CLR(d, -1);
d[s] = 0;
Q.push(s);
while (!Q.empty())
{
int u = Q.front();
Q.pop();
for (int i = head[u]; ~i; i = E[i].nxt)
{
int v = E[i].to;
if (d[v] == -1 && E[i].cap > 0)
{
d[v] = d[u] + 1;
if (v == t)
return 1;
Q.push(v);
}
}
}
return ~d[t];
}
double dfs(int s, int t, double f)
{
if (s == t || f == 0)
return f;
double ret = 0;
for (int i = head[s]; ~i; i = E[i].nxt)
{
int v = E[i].to;
if (d[v] == d[s] + 1 && E[i].cap > 0)
{
double df = dfs(v, t, min(f, E[i].cap));
if (df > 0)
{
E[i].cap -= df;
E[i ^ 1].cap += df;
ret += df;
f -= df;
if (f == 0)
break;
}
}
}
if (ret == 0)
d[s] = -1;
return ret;
}
double dinic(int s, int t)
{
double ret = 0;
while (bfs(s, t))
ret += dfs(s, t, 1e9);
return ret;
}
double CST(int n, int m, double g)
{
init();
double U = m;
int S = 0, T = n + 1;
int i;
for (i = 1; i <= n; ++i) //<S,v,U>与<v,T,U+2g-dv>
{
add(S, i, U);
add(i, T, U + 2 * g - 1.0 * deg[i]);
}
for (i = 1; i <= m; ++i)//(u,v,1)
{
add(e[i].u, e[i].v, 1.0);
add(e[i].v, e[i].u, 1.0);
}
return dinic(S, T); //直接计算最小割值(最大流值)
}
int main(void)
{
int n, m, i;
while (~scanf("%d%d", &n, &m))
{
CLR(deg, 0);
for (i = 1; i <= m; ++i)
{
scanf("%d%d", &e[i].u, &e[i].v);
++deg[e[i].u];
++deg[e[i].v];
}
if (!m)
{
puts("1");
puts("1");
}
else
{
double L = 1.0 / n, R = m;
double ans = 1;
double dx = 1.0 / n / n;
while (fabs(R - L) >= dx)
{
double mid = (L + R) / 2.0;
double hg = (m * n * 1.0 - CST(n, m, mid)) / 2;
if (hg >= eps)
{
L = mid;
ans = mid;
}
else
R = mid;
}
CST(n, m, ans);
vector<int>pos;
for (i = 1; i <= n; ++i)
if (~d[i])
pos.push_back(i);
int sz = pos.size();
sort(pos.begin(), pos.end());
printf("%d\n", sz);
for (i = 0; i < sz; ++i)
printf("%d\n", pos[i]);
}
}
return 0;
}

POJ 3155 Hard Life(最大密度子图+改进算法)的更多相关文章

  1. POJ 3155 Hard Life 最大密度子图 最大权闭合图 网络流 二分

    http://poj.org/problem?id=3155 最大密度子图和最大权闭合图性质很相近(大概可以这么说吧),一个是取最多的边一个是取最多有正贡献的点,而且都是有选一种必须选另一种的限制,一 ...

  2. POJ 3155 Hard Life(最大密度子图)

    裸题.输入一个无向图,输出最大密度子图(输出子图结点数和升序编号). 看了<最小割模型在信息学竞赛中的应用——胡伯涛>的一部分,感觉01分数规划问题又是个大坑.暂时还看不懂. 参考http ...

  3. poj 3155 最大密度子图

    思路: 这个还是看的胡伯涛的论文<最小割在信息学竞赛中的应用>.是将最大密度子图问题转化为了01分数规划和最小割问题. 直接上代码: #include <iostream> # ...

  4. POJ 3155:Hard Life(最大密度子图)

    题目链接 题意 给出n个人,和m对有冲突的人.要裁掉一些人,使得冲突率最高,冲突率为存在的冲突数/人数. 思路 题意可以转化为,求出一些边,使得|E|/|V|最大,这种分数规划叫做最大密度子图. 学习 ...

  5. 静态频繁子图挖掘算法用于动态网络——gSpan算法研究

    摘要 随着信息技术的不断发展,人类可以很容易地收集和储存大量的数据,然而,如何在海量的数据中提取对用户有用的信息逐渐地成为巨大挑战.为了应对这种挑战,数据挖掘技术应运而生,成为了最近一段时期数据科学的 ...

  6. POJ 3155 Hard Life

    Hard Life Time Limit: 8000ms Memory Limit: 65536KB This problem will be judged on PKU. Original ID:  ...

  7. 排序系列 之 简单选择排序及其改进算法 —— Java实现

    简单选择排序算法: 基本思想: 在待排序数据中,选出最小的一个数与第一个位置的数交换:然后在剩下的数中选出最小的数与第二个数交换:依次类推,直至循环到只剩下两个数进行比较为止. 实例: 0.初始状态 ...

  8. POJ3155 Hard Life [最大密度子图]

      题意:最大密度子图 #include<iostream> #include<cstdio> #include<cstring> #include<algo ...

  9. KMP及其改进算法

    本文主要讲述KMP已经KMP的一种改进方法.若发现不正确的地方,欢迎交流指出,谢谢! KMP算法的基本思想: KMP的算法流程: 每当一趟匹配过程中出现字符比较不等时,不需回溯 i 指针,而是利用已经 ...

随机推荐

  1. 2018.5.29 Oracle连接到空闲例程

    解决方法如下: 1.通过cmd命令窗启动Oracle:(最好是以管理员身份启动) C:\Users\Administrator>sqlplus /nolog SQL>conn /as sy ...

  2. imfilter()用法

    功能:对图像进行滤波. 用法: g = imfilter(f, w, filtering_mode, boundary_options, size_options) 其中,f:输入图像,w:滤波掩模, ...

  3. DeepLearning tutorial(3)MLP多层感知机原理简介+代码详解

    本文介绍多层感知机算法,特别是详细解读其代码实现,基于python theano,代码来自:Multilayer Perceptron,如果你想详细了解多层感知机算法,可以参考:UFLDL教程,或者参 ...

  4. 使用dotnet-dump 查找 .net core 3.0 占用CPU 100%的原因

    公司的产品一直紧跟 .net core 3.0 preview 不断升级, 部署到 Linux 服务器后, 偶尔会出现某个进程CPU占用100%. 由于服务部署在云上, 不能使用远程调试; 在局域网内 ...

  5. shell脚本,awk结合正则来打印文件里面的内容。

    文件内容如下:key1abc d key2 1.想得到如下结果: abc d 2.想得到如下结果: key1key2

  6. FILE对象线程安全

    根据apue讲述: 标准的IO例程可能从它们各自的内部数据结构的角度出发,是以线程安全的方式实现的!但在线程中,如果标准 IO例程都获取它们各自的锁,那么在做一次一个字符的IO时就会出现严重的性能下降 ...

  7. C/C++程序基础 (九)排序算法简述

    排序算法 算法复杂度 算法简述 插入排序 N2 前方有序,依次将后方无序数据插入前方合适位置. 冒泡排序 N2 前方有序,从后方两两比较,将最小泡冒到前方. 选择排序 N2 前方有序,从后方选择最小的 ...

  8. 【Python学习之八】ORM

    ORM 什么是ORM呢? ORM全称是:Object-Relational Mapping.即对象-关系映射,就是把关系数据库的一行映射为一个对象,也就是一个类对应一个表.这样,写代码更简单,不用直接 ...

  9. 【java】A local class access to local variables

    内部类参考 A local class has access to local variables. However, a local class can only access local vari ...

  10. linux 基本指令 归类

    今天 我们来学习一下 最最基础的linux 指令,在我看来 linux的操作就是 增 删 改 查 这四个字. 1 查询 操作用户 woami 2查询登录用户 who am i 2 pwd //查询当前 ...