【BZOJ-3638&3272&3267&3502】k-Maximum Subsequence Sum 费用流构图 + 线段树手动增广
3638: Cf172 k-Maximum Subsequence Sum
Time Limit: 50 Sec Memory Limit: 256 MB
Submit: 174 Solved: 92
[Submit][Status][Discuss]
Description
给一列数,要求支持操作: 1.修改某个数的值 2.读入l,r,k,询问在[l,r]内选不相交的不超过k个子段,最大的和是多少。
Input
The first line contains integer n (1 ≤ n ≤ 105), showing how many numbers the sequence has. The next line contains n integers a1, a2, ..., an (|ai| ≤ 500).
The third line contains integer m (1 ≤ m ≤ 105) — the number of queries. The next m lines contain the queries in the format, given in the statement.
All changing queries fit into limits: 1 ≤ i ≤ n, |val| ≤ 500.
All queries to count the maximum sum of at most k non-intersecting subsegments fit into limits: 1 ≤ l ≤ r ≤ n, 1 ≤ k ≤ 20. It is guaranteed that the number of the queries to count the maximum sum of at mostk non-intersecting subsegments doesn't exceed 10000.
Output
For each query to count the maximum sum of at most k non-intersecting subsegments print the reply — the maximum sum. Print the answers to the queries in the order, in which the queries follow in the input.
Sample Input
9 -8 9 -1 -1 -1 9 -8 9
3
1 1 9 1
1 1 9 2
1 4 6 3
Sample Output
25
0
HINT
In the first query of the first example you can select a single pair (1, 9). So the described sum will be 17.
Look at the second query of the first example. How to choose two subsegments? (1, 3) and (7, 9)? Definitely not, the sum we could get from (1, 3) and (7, 9) is 20, against the optimal configuration (1, 7) and (9, 9) with 25.
The answer to the third query is 0, we prefer select nothing if all of the numbers in the given interval are negative.
Source
3272: Zgg吃东西
Time Limit: 10 Sec Memory Limit: 256 MB
Submit: 93 Solved: 59
[Submit][Status][Discuss]
Description
Input
Output
Sample Input
9 -8 9 -1 -1 -1 9 -8 9
5
1 1 9 1
1 1 9 2
1 4 6 3
0 3 -8
1 1 9 1
Sample Output
25
0
10
HINT
对于100%的数据 N,M<=100000 ,1<=k<=20.l<=l<=r<=n.数值的绝对值不会超过500.
Zgg是可以有手空闲的…
Source
3267: KC采花
Time Limit: 30 Sec Memory Limit: 256 MB
Submit: 65 Solved: 51
[Submit][Status][Discuss]
Description
Input
Output
Sample Input
9 -8 9 -1 -1 -1 9 -8 9
3
1 1 9 1
1 1 9 2
1 4 6 3
Sample Output
25
0
HINT
100%的数据,满足n,m<=50000,k<=20,ai以及修改的val的绝对值不超过500。
Source
Solution
这道题,首先想到最大费用最大流,这类的对序列上的问题,是比较经典的问题,详见 《网络流与线性规划24题》最长k可重区间集问题
那么这道题,最标准的做法就是:
把每个数拆成两个点$I_{1}$和$I_{2}$ $I_{1}-->I_{2}$,$cap=1$,$cost=a[i]$
然后$S-->I_{1}$,$cap=INF$,$cost=0$;以及$I_{2}-->T$,$cap=INF$,$cost=0$;
并且相邻的两位连$I'_{2}-->I''_{1}$,$cap=1$,$cost=0$
那么只需要跑$K$次最大费用最大流,累加答案即可
但是很显然这种数据范围,这么做会TLE,那么不妨考虑对这种做法进行优化:
每次增广的过程,实质上就是取一段和最大的子序列,并将其反转
很显然对于这类序列上的操作,可以用线段树去实现,那么这道题的正确做法就明确了
费用流的构图,用线段树去手动模拟增广路程
线段树维护的方法,只需要维护一段区间的最大和子序列,涉及取反,所以还需要维护最小和子序列
但是发现,每进行一次取反,当前最大和子序列一定变成最小和子序列,最小和子序列一定变成最大,那么直接swap一下就好
鉴于一次询问需要增广K次,也就是一次询问要取反,所以显然需要开一个栈记录一下当前询问所反转的所有区间,在结束时还原
修改操作直接修改即可
总的时间复杂度是$O(knlogn)$
Code
#include<iostream>
#include<cmath>
#include<cstring>
#include<algorithm>
#include<cstdio>
using namespace std;
inline int read()
{
int x=,f=; char ch=getchar();
while (ch<'' || ch>'') {if (ch=='-') f=-; ch=getchar();}
while (ch>='' && ch<='') {x=x*+ch-''; ch=getchar();}
return x*f;
}
#define maxn 101000
int n,m,a[maxn],ans;
struct DataNode
{
int lx,rx,mx,sum,pl,pr,p1,p2;
void add(int pos,int x) {p1=p2=pl=pr=pos,mx=sum=lx=rx=x;}
};
struct SegmentTreeNode
{
int l,r,tag;
DataNode maxx,minn;
void add(int x) {maxx.add(l,x); minn.add(l,-x);}
}tree[maxn<<];
inline DataNode Merge(DataNode ls,DataNode rs)
{
DataNode rt;
rt.sum=ls.sum+rs.sum;
rt.lx=ls.sum+rs.lx>ls.lx? ls.sum+rs.lx : ls.lx;
rt.pl=ls.sum+rs.lx>ls.lx? rs.pl : ls.pl;
rt.rx=rs.sum+ls.rx>rs.rx? rs.sum+ls.rx : rs.rx;
rt.pr=rs.sum+ls.rx>rs.rx? ls.pr : rs.pr;
rt.mx=ls.rx+rs.lx; rt.p1=ls.pr; rt.p2=rs.pl;
if (rt.mx<ls.mx) rt.mx=ls.mx,rt.p1=ls.p1,rt.p2=ls.p2;
if (rt.mx<rs.mx) rt.mx=rs.mx,rt.p1=rs.p1,rt.p2=rs.p2;
return rt;
}
inline void Update(int now)
{
tree[now].minn=Merge(tree[now<<].minn,tree[now<<|].minn);
tree[now].maxx=Merge(tree[now<<].maxx,tree[now<<|].maxx);
}
inline void Pushdown(int now)
{
if (!tree[now].tag || tree[now].l==tree[now].r) return;
tree[now].tag^=;
swap(tree[now<<].maxx,tree[now<<].minn);
swap(tree[now<<|].minn,tree[now<<|].maxx);
tree[now<<].tag^=; tree[now<<|].tag^=;
}
inline void BuildTree(int now,int l,int r)
{
tree[now].l=l; tree[now].r=r; tree[now].tag=;
if (l==r) {tree[now].add(a[l]); return;}
int mid=(l+r)>>;
BuildTree(now<<,l,mid); BuildTree(now<<|,mid+,r);
Update(now);
}
inline void Change(int now,int pos,int x)
{
Pushdown(now);
if (tree[now].l==tree[now].r) {tree[now].add(x); return;}
int mid=(tree[now].l+tree[now].r)>>;
if (pos<=mid) Change(now<<,pos,x); else Change(now<<|,pos,x);
Update(now);
}
inline DataNode Query(int now,int L,int R)
{
Pushdown(now);
if (L==tree[now].l && R==tree[now].r) return tree[now].maxx;
int mid=(tree[now].l+tree[now].r)>>;
if (R<=mid) return Query(now<<,L,R);
if (L>mid) return Query(now<<|,L,R);
return Merge(Query(now<<,L,mid),Query(now<<|,mid+,R));
}
inline void Reverse(int now,int L,int R)
{
Pushdown(now);
if (L<=tree[now].l && R>=tree[now].r)
{swap(tree[now].maxx,tree[now].minn); tree[now].tag^=; return;}
int mid=(tree[now].l+tree[now].r)>>;
if (L<=mid) Reverse(now<<,L,R);
if (R>mid) Reverse(now<<|,L,R);
Update(now);
}
struct StackNode
{
int l,r;
void Push(int L,int R) {l=L,r=R;}
}stack[];
int top;
inline void SolveAsk(int L,int R,int K)
{
ans=; top=;
while (K--)
{
DataNode tmp=Query(,L,R);
if (tmp.mx>) ans+=tmp.mx; else break;
Reverse(,tmp.p1,tmp.p2);
stack[++top].Push(tmp.p1,tmp.p2);
}
for (int i=top; i>=; i--)
Reverse(,stack[i].l,stack[i].r);
printf("%d\n",ans);
}
inline void SolveChange(int pos,int x) {Change(,pos,x);}
void Freopen() {freopen("data.in","r",stdin);freopen("data.out","w",stdout);}
void Fclose() {fclose(stdin);fclose(stdout);}
int main()
{
// Freopen();
n=read();
for (int i=; i<=n; i++) a[i]=read();
BuildTree(,,n);
m=read();
while (m--)
{
int opt=read(),L,R,K,x,d;
switch (opt)
{
case : L=read(),R=read(),K=read(); SolveAsk(L,R,K); break;
case : x=read(),d=read(); SolveChange(x,d); break;
}
}
// Fclose();
return ;
}
3502: PA2012 Tanie linie
Time Limit: 10 Sec Memory Limit: 256 MB
Submit: 181 Solved: 36
[Submit][Status][Discuss]
Description
n个数字,求不相交的总和最大的最多k个连续子序列。
1<= k<= N<= 1000000。
Input
Output
Sample Input
7 -3 4 -9 5
Sample Output
HINT
Source
Solution
和刚刚一样,但是范围扩大了,需要开longlong,所以内存很卡,时间也很卡..
Code
#include<iostream>
#include<cmath>
#include<cstring>
#include<algorithm>
#include<cstdio>
using namespace std;
inline int read()
{
int x=,f=; char ch=getchar();
while (ch<'' || ch>'') {if (ch=='-') f=-; ch=getchar();}
while (ch>='' && ch<='') {x=x*+ch-''; ch=getchar();}
return x*f;
}
#define maxn 1000010
int n,m,a[maxn];long long ans;
struct DataNode
{
long long lx,rx,mx,sum;int pl,pr,p1,p2;
void add(int pos,int x) {p1=p2=pl=pr=pos,mx=sum=lx=rx=x;}
};
struct SegmentTreeNode
{
int l,r,tag;
DataNode maxx,minn;
void add(long long x) {maxx.add(l,x); minn.add(l,-x);}
}tree[maxn*+];
inline DataNode Merge(DataNode ls,DataNode rs)
{
DataNode rt;
rt.sum=ls.sum+rs.sum;
rt.lx=ls.sum+rs.lx>ls.lx? ls.sum+rs.lx : ls.lx;
rt.pl=ls.sum+rs.lx>ls.lx? rs.pl : ls.pl;
rt.rx=rs.sum+ls.rx>rs.rx? rs.sum+ls.rx : rs.rx;
rt.pr=rs.sum+ls.rx>rs.rx? ls.pr : rs.pr;
rt.mx=ls.rx+rs.lx; rt.p1=ls.pr; rt.p2=rs.pl;
if (rt.mx<ls.mx) rt.mx=ls.mx,rt.p1=ls.p1,rt.p2=ls.p2;
if (rt.mx<rs.mx) rt.mx=rs.mx,rt.p1=rs.p1,rt.p2=rs.p2;
return rt;
}
inline void Update(int now)
{
tree[now].minn=Merge(tree[now<<].minn,tree[now<<|].minn);
tree[now].maxx=Merge(tree[now<<].maxx,tree[now<<|].maxx);
}
inline void Pushdown(int now)
{
if (!tree[now].tag || tree[now].l==tree[now].r) return;
tree[now].tag^=;
swap(tree[now<<].maxx,tree[now<<].minn);
swap(tree[now<<|].minn,tree[now<<|].maxx);
tree[now<<].tag^=; tree[now<<|].tag^=;
}
inline void BuildTree(int now,int l,int r)
{
tree[now].l=l; tree[now].r=r; tree[now].tag=;
if (l==r) {tree[now].add(a[l]); return;}
int mid=(l+r)>>;
BuildTree(now<<,l,mid); BuildTree(now<<|,mid+,r);
Update(now);
}
inline DataNode Query(int now,int L,int R)
{
Pushdown(now);
if (L==tree[now].l && R==tree[now].r) return tree[now].maxx;
int mid=(tree[now].l+tree[now].r)>>;
if (R<=mid) return Query(now<<,L,R);
if (L>mid) return Query(now<<|,L,R);
return Merge(Query(now<<,L,mid),Query(now<<|,mid+,R));
}
inline void Reverse(int now,int L,int R)
{
Pushdown(now);
if (L<=tree[now].l && R>=tree[now].r)
{swap(tree[now].maxx,tree[now].minn); tree[now].tag^=; return;}
int mid=(tree[now].l+tree[now].r)>>;
if (L<=mid) Reverse(now<<,L,R);
if (R>mid) Reverse(now<<|,L,R);
Update(now);
}
inline void SolveAsk(int L,int R,int K)
{
while (K--)
{
DataNode tmp=Query(,L,R);
if (tmp.mx>) ans+=tmp.mx; else break;
Reverse(,tmp.p1,tmp.p2);
}
printf("%lld\n",ans);
}
int main()
{
n=read(); int K=read();
for (int i=; i<=n; i++) a[i]=read();
BuildTree(,,n);
SolveAsk(,n,K);
return ;
}
【BZOJ-3638&3272&3267&3502】k-Maximum Subsequence Sum 费用流构图 + 线段树手动增广的更多相关文章
- BZOJ 3836 Codeforces 280D k-Maximum Subsequence Sum (模拟费用流、线段树)
题目链接 (BZOJ) https://www.lydsy.com/JudgeOnline/problem.php?id=3836 (Codeforces) http://codeforces.com ...
- BZOJ 5326 [JSOI2017]博弈 (模拟费用流、线段树)
题目链接 https://www.lydsy.com/JudgeOnline/problem.php?id=5326 题解 终于成为第8个A掉这题的人--orz tzw神仙早我6小时 本以为这东西常数 ...
- Codeforces 280D k-Maximum Subsequence Sum [模拟费用流,线段树]
洛谷 Codeforces bzoj1,bzoj2 这可真是一道n倍经验题呢-- 思路 我首先想到了DP,然后矩阵,然后线段树,然后T飞-- 搜了题解之后发现是模拟费用流. 直接维护选k个子段时的最优 ...
- BZOJ 1920 Luogu P4217 [CTSC2010]产品销售 (模拟费用流、线段树)
题目链接 (bzoj) https://www.lydsy.com/JudgeOnline/problem.php?id=1920 (luogu) https://www.luogu.org/prob ...
- Maximum Subsequence Sum【最大连续子序列+树状数组解决】
Problem Description 给定K个整数的序列{ N1, N2, ..., NK },其任意连续子序列可表示为{ Ni, Ni+1, ..., Nj },其中 1 <= i < ...
- 中国大学MOOC-陈越、何钦铭-数据结构-2015秋 01-复杂度2 Maximum Subsequence Sum (25分)
01-复杂度2 Maximum Subsequence Sum (25分) Given a sequence of K integers { N1,N2, ..., NK }. ...
- PAT1007:Maximum Subsequence Sum
1007. Maximum Subsequence Sum (25) 时间限制 400 ms 内存限制 65536 kB 代码长度限制 16000 B 判题程序 Standard 作者 CHEN, Y ...
- PTA (Advanced Level) 1007 Maximum Subsequence Sum
Maximum Subsequence Sum Given a sequence of K integers { N1, N2, ..., NK }. A continuous su ...
- 【DP-最大子串和】PAT1007. Maximum Subsequence Sum
1007. Maximum Subsequence Sum (25) 时间限制 400 ms 内存限制 32000 kB 代码长度限制 16000 B 判题程序 Standard 作者 CHEN, Y ...
随机推荐
- php中的ip2long和long2ip的理解
IPv4地址是如何表示的 IPv4使用无符号32位地址,因此最多有2的32次方减1(4294967295)个地址.一般的书写法为用4个小数点分开的十进制数,记为:A.B.C.D,比如:157.23.5 ...
- Centos下Tomcat 安装Apache Portable Runtime
APR(Apache Portable Runtime)是一个高可移植库,它是Apache HTTP Server 2.x的核心. APR有很多用途,包括访问高级IO功能(例如sendfile,epo ...
- 02传智_jbpm与OA项目_部门模块
部门模块:具有增删改查(部门)的功能. Dao层的实现: 1,定义一个DepartmentDao.java. 定义基本的数据库操作.
- Linux wait函数详解
wait和waitpid出现的原因 SIGCHLD --当子进程退出的时候,内核会向父进程SIGCHLD信号,子进程的退出是个异步事件(子进程可以在父进程运行的任何时刻终止) --子进程退出时,内核将 ...
- 【语言基础】c++ 基本数据类型与字节数组(string,char [] )之间的转化方法
有时候我们需要将基本数据类型转化为字节,以便写入文件,然后必要时还需要将这些字节读出来.有人说,为啥不把数字直接存进文件呢?比如:100,000,000,我们直接存数字明文到文件那就是9个字符(cha ...
- spring boot/cloud 应用监控
应用的监控功能,对于分布式系统非常重要.如果把分布式系统比作整个社会系统.那么各个服务对应社会中具体服务机构,比如银行.学校.超市等,那么监控就类似于警察局和医院,所以其重要性显而易见.这里说的,监控 ...
- 171 Excel Sheet Column Number
/** * 题意:A表示1 B表示2 AA表示27 AB表示28 ------>给你一串字符串输出相应的数字 * 分析:这个就类似于二进制转十进制,从字符串后面往前遍历,然后pow(26,n)* ...
- linux的点滴积累
p.p1 { margin: 0.0px 0.0px 0.0px 0.0px; font: 18.0px Menlo; color: #839496; background-color: rgba(1 ...
- Learning to Rank 之 listwise ranking
详细文章: http://www.machinelearning.org/proceedings/icml2007/papers/139.pdf
- STL中的next_permutation
给定一个数组a[N],求下一个数组. 2 1 3 4 2 1 4 3 2 3 1 4 2 3 4 1 ..... 在STL中就有这个函数: 1.参数是(数组的第一个元素,数组的末尾),注意这是前闭后开 ...