D. Sum of Medians
time limit per test

3 seconds

memory limit per test

256 megabytes

input

standard input

output

standard output

In one well-known algorithm of finding the k-th order statistics we should divide all elements into groups of five consecutive elements and find the median of each five. A median is called the middle element of a sorted array (it's the third largest element for a group of five). To increase the algorithm's performance speed on a modern video card, you should be able to find a sum of medians in each five of the array.

A sum of medians of a sorted k-element set S = {a1, a2, ..., ak}, where a1 < a2 < a3 < ... < ak, will be understood by as

The  operator stands for taking the remainder, that is  stands for the remainder of dividing x by y.

To organize exercise testing quickly calculating the sum of medians for a changing set was needed.

Input

The first line contains number n (1 ≤ n ≤ 105), the number of operations performed.

Then each of n lines contains the description of one of the three operations:

  • add x — add the element x to the set;
  • del x — delete the element x from the set;
  • sum — find the sum of medians of the set.

For any add x operation it is true that the element x is not included in the set directly before the operation.

For any del x operation it is true that the element x is included in the set directly before the operation.

All the numbers in the input are positive integers, not exceeding 109.

Output

For each operation sum print on the single line the sum of medians of the current set. If the set is empty, print 0.

Please, do not use the %lld specificator to read or write 64-bit integers in C++. It is preferred to use the cin, cout streams (also you may use the %I64d specificator).

Sample test(s)
input
6
add 4
add 5
add 1
add 2
add 3
sum
output
3
input
14
add 1
add 7
add 2
add 5
sum
add 6
add 8
add 9
add 3
add 4
add 10
sum
del 1
sum
output

5
11
13
-----------------------------------------------------------------------------------------------

Solution
用线段树维护集合
1.预处理所有add/del操作,将涉及的元素离散化为1...n
2.add x ,设x离散化为i,修改区间(i+1, n), 将各节点的sum[5]右移一位,再插入x
3.del x, 设x离散化为i, 修改区间(i+1, n), 将各节点的sum[5]左移一位, 在删除x
4.插入/删除元素需要知道元素在集合中的位置 (place),这由树状数组(BIT)维护。
5.对区间sum[5]的左移/右移操作需要用lazy-tag优化。
6.如果手写离散化,要注意实现细节。
7.需考虑极端输入,比如无sum询问的输入
---------------------------------------------------------------------------------------------
这是我第一发提交,TLE on test 5, 跪在极端输入上了,TLE的原因是调用了add(0, 1)
#include<bits/stdc++.h>
using namespace std;
typedef long long ll;
const int MAX_N=1e5+;
#define X first
#define Y second
//BIT
int bit[MAX_N], rb;
int sum(int i){
int s=;
while(i){
s+=bit[i];
i-=i&-i;
}
return s;
}
void add(int i, int x){
while(i<=rb){
bit[i]+=x;
i+=i&-i;
}
} typedef pair<int, int> P;
P q[MAX_N];
int val[MAX_N];
char type[MAX_N][];
int lisan(int N){
int v;
for(int i=; i<N; i++){
scanf("%s", type[i]);
switch(*type[i]){
case 'a':
case 'd':
scanf("%d", &v);
q[i]=P(v, i);
break;
case 's':
q[i]=P(, i);  //TLE在这里
break;
}
}
sort(q, q+N);
int ord=;
for(int i=; i<N; ord++){
val[ord]=q[i].X;
do{
q[i].X=ord;
i++;
}while(i<N&&q[i].X==val[ord]);
}
return ord;
}
bool cmp(const P &a, const P &b){
return a.Y<b.Y;
}
//ST
struct node{
int l, r, tag;
ll sum[];
int mid(){return (l+r)>>;}
}T[MAX_N<<];
void shift(int id, int cnt){
cnt%=;
if(!cnt) return;
ll tmp[];
memcpy(tmp, T[id].sum, sizeof(tmp));
for(int i=; i<; i++){
int nt=(i+cnt+)%;
T[id].sum[nt]=tmp[i];
}
}
void pushdown(int id){
node &now=T[id], &lch=T[id<<], &rch=T[id<<|];
lch.tag+=now.tag, rch.tag+=now.tag;
shift(id<<, now.tag);
shift(id<<|, now.tag);
now.tag=; //error prone
}
void unite(int id){
node &now=T[id], &lch=T[id<<], &rch=T[id<<|];
for(int i=; i<; i++)
now.sum[i]=lch.sum[i]+rch.sum[i];
}
void build(int id, int l, int r){
node &now=T[id];
T[id].l=l, T[id].r=r, T[id].tag=;
memset(T[id].sum, , sizeof(T[id].sum));
if(l==r) return;
int mid=(l+r)>>;
build(id<<, l, mid);
build(id<<|, mid+, r);
}
void insert(int id, int pos, int ord, int v){
node &now=T[id];
if(now.l==now.r){
now.sum[ord%]+=v;
}
else{
pushdown(id);
if(pos<=now.mid())
insert(id<<, pos, ord, v);
else
insert(id<<|, pos, ord, v);
unite(id);
}
}
void shift(int id, int lb, int cnt){
if(lb>rb) return;
node &now=T[id];
if(now.l>=lb){
shift(id, cnt);
now.tag+=cnt;
}
else{
pushdown(id);
if(lb<=now.mid())
shift(id<<, lb, cnt);
shift(id<<|, lb, cnt);
unite(id);
}
} int main(){
//freopen("in", "r", stdin);
int N;
scanf("%d", &N);
rb=lisan(N);
build(, , rb);
sort(q, q+N, cmp);
for(int i=; i<N; i++){
if(*type[i]=='a'){
int ord=sum(q[i].X)+;
add(q[i].X, );
shift(, q[i].X+, );
insert(, q[i].X, ord, val[q[i].X]);
}
else if(*type[i]=='d'){
int ord=sum(q[i].X);
add(q[i].X, -);
shift(, q[i].X+, -);
insert(, q[i].X, ord, -val[q[i].X]);
}
else{
printf("%I64d\n", T[].sum[]);
}
}
return ;
}

其实我已经考虑到这种无sum询问的输入,但没注意到它会导致我的代码TLE

AC version

#include<bits/stdc++.h>
using namespace std;
typedef long long ll;
const int MAX_N=1e5+;
#define X first
#define Y second
//BIT
int bit[MAX_N], rb;
int sum(int i){
int s=;
while(i){
s+=bit[i];
i-=i&-i;
}
return s;
}
void add(int i, int x){
while(i<=rb){
bit[i]+=x;
i+=i&-i;
}
} typedef pair<int, int> P;
P q[MAX_N];
int val[MAX_N];
char type[MAX_N][];
int lisan(int N){
int v;
for(int i=; i<N; i++){
scanf("%s", type[i]);
switch(*type[i]){
case 'a':
case 'd':
scanf("%d", &v);
q[i]=P(v, i);
break;
case 's':
q[i]=P(INT_MAX, i);  //注意这里的改动
break;
}
}
sort(q, q+N);
int ord=;
for(int i=; i<N;){
++ord;
val[ord]=q[i].X;
do{
q[i].X=ord;
i++;
}while(i<N&&q[i].X==val[ord]);
}
return ord;
}
bool cmp(const P &a, const P &b){
return a.Y<b.Y;
}
//ST
struct node{
int l, r, tag;
ll sum[];
int mid(){return (l+r)>>;}
}T[MAX_N<<];
void shift(int id, int cnt){
cnt%=;
if(!cnt) return;
ll tmp[];
memcpy(tmp, T[id].sum, sizeof(tmp));
for(int i=; i<; i++){
int nt=(i+cnt+)%;
T[id].sum[nt]=tmp[i];
}
}
void pushdown(int id){
node &now=T[id], &lch=T[id<<], &rch=T[id<<|];
lch.tag+=now.tag, rch.tag+=now.tag;
shift(id<<, now.tag);
shift(id<<|, now.tag);
now.tag=; //error prone
}
void unite(int id){
node &now=T[id], &lch=T[id<<], &rch=T[id<<|];
for(int i=; i<; i++)
now.sum[i]=lch.sum[i]+rch.sum[i];
}
void build(int id, int l, int r){
node &now=T[id];
T[id].l=l, T[id].r=r, T[id].tag=;
memset(T[id].sum, , sizeof(T[id].sum));
if(l==r) return;
int mid=(l+r)>>;
build(id<<, l, mid);
build(id<<|, mid+, r);
}
void insert(int id, int pos, int ord, int v){
node &now=T[id];
if(now.l==now.r){
now.sum[ord%]+=v;
}
else{
pushdown(id);
if(pos<=now.mid())
insert(id<<, pos, ord, v);
else
insert(id<<|, pos, ord, v);
unite(id);
}
}
void shift(int id, int lb, int cnt){
if(lb>rb) return;
node &now=T[id];
if(now.l>=lb){
shift(id, cnt);
now.tag+=cnt;
}
else{
pushdown(id);
if(lb<=now.mid())
shift(id<<, lb, cnt);
shift(id<<|, lb, cnt);
unite(id);
}
} int main(){
//freopen("in", "r", stdin);
int N;
scanf("%d", &N);
rb=lisan(N);
build(, , rb);
sort(q, q+N, cmp);
for(int i=; i<N; i++){
if(*type[i]=='a'){
int ord=sum(q[i].X)+;
add(q[i].X, );
shift(, q[i].X+, );
insert(, q[i].X, ord, val[q[i].X]);
}
else if(*type[i]=='d'){
int ord=sum(q[i].X);
add(q[i].X, -);
shift(, q[i].X+, -);
insert(, q[i].X, ord, -val[q[i].X]);
}
else{
printf("%I64d\n", T[].sum[]);
}
}
return ;
}

Codeforces 85D Sum of Medians的更多相关文章

  1. Codeforces 85D Sum of Medians(线段树)

    题目链接:Codeforces 85D - Sum of Medians 题目大意:N个操作,add x:向集合中加入x:del x:删除集合中的x:sum:将集合排序后,将集合中全部下标i % 5 ...

  2. 数据结构(线段树):CodeForces 85D Sum of Medians

    D. Sum of Medians time limit per test 3 seconds memory limit per test 256 megabytes input standard i ...

  3. CodeForces 85D Sum of Medians Splay | 线段树

    Sum of Medians 题解: 对于这个题目,先想到是建立5棵Splay,然后每次更新把后面一段区间的树切下来,然后再转圈圈把切下来的树和别的树合并. 但是感觉写起来太麻烦就放弃了. 建立5棵线 ...

  4. 85D Sum of Medians

    传送门 题目 In one well-known algorithm of finding the k-th order statistics we should divide all element ...

  5. CF 85D Sum of Medians (五颗线段树)

    http://codeforces.com/problemset/problem/85/D 题意: 给你N(0<N<1e5)次操作,每次操作有3种方式, 1.向集合里加一个数a(0< ...

  6. codeforces 85D D. Sum of Medians Vector的妙用

    D. Sum of Medians Time Limit: 1 Sec  Memory Limit: 256 MB 题目连接 http://codeforces.com/problemset/prob ...

  7. codeforces 85D D. Sum of Medians 线段树

    D. Sum of Medians time limit per test 3 seconds memory limit per test 256 megabytes input standard i ...

  8. Yandex.Algorithm 2011 Round 1 D. Sum of Medians 线段树

    题目链接: Sum of Medians Time Limit:3000MSMemory Limit:262144KB 问题描述 In one well-known algorithm of find ...

  9. Coderforces 85 D. Sum of Medians(线段树单点修改)

    D. Sum of Medians time limit per test 3 seconds memory limit per test 256 megabytes input standard i ...

随机推荐

  1. SAP中主数据和单据的删除

    在SAP实际操作的过程中,有些主数据或者单据需要删除,但是删除的方法却不尽相同,所以笔者今天总结了下,供大家参考. 1,用T-Code去删除 例如我们要删除某个物料,我们可以用T-Code MM06 ...

  2. Volley(六 )—— 从源码带看Volley的缓存机制

    磁盘缓存DiskBasedCache 如果你还不知道volley有磁盘缓存的话,请看一下我的另一篇博客请注意,Volley已默认使用磁盘缓存 DiskBasedCache内部结构 它由两部分组成,一部 ...

  3. SVN和git的使用(附github的简单玩法)

    今天简单的总结了下SVN和git的使用,也尝试了下github,应该好好提高下自己的英文水平了,梦想有一天不再使用任何翻译软件. [svn]:集中式的代码管理工具(版本控制工具--版本记录) 1> ...

  4. Gitub

    1.下载地址(注册:jackchn,jackchn@foxmail.com) http://windows.github.com/ 2.使用 github for Windows使用介绍 搭建一个免费 ...

  5. wireshark排查打印机问题

    抓包工具排除故障 前言:上网慢,可能是内网堵了.装上wireshark,可抓到广播包,多播包,以及发给自己的包.如果想抓lan内其他人之间的通信包,那就要在sw上做端口镜像. 背景 调试打印机的人发现 ...

  6. pandas 给数据打标签

    import numpy as np import pandas as pd df = pd.DataFrame(np.random.randint(0,100,100), columns=['sco ...

  7. iOS 定位精度

    时间 2015-03-19 18:30:59  图灵社区  由于iOS不能直接控制到GPS,一般来说我们都会使用CLLocationManager来获取地理位置信息,我们会使用 manager.des ...

  8. 20135220谈愈敏Linux Book_17

    第17章 设备与模块 关于设备驱动和设备管理的四种内核成分: 设备类型:在所有 Unix 系统中为了统一普通设备的操作所采用的分类. 模块: Linux 内核中用于按需加载和卸载目标码的机制. 内核对 ...

  9. 信息安全系统设计基础exp_5

    北京电子科技学院(BESTI) 实     验    报     告 课程:信息安全系统设计基础 班级:1353 姓名:郑伟.吴子怡 学号:20135322.20135313 指导教师: 娄嘉鹏 实验 ...

  10. 在茫茫人海中发现相似的你——局部敏感哈希(LSH)

    一.引入 在做微博文本挖掘的时候,会发现很多微博是高度相似的,因为大量的微博都是转发其他人的微博,并且没有添加评论,导致很多数据是重复或者高度相似的.这给我们进行数据处理带来很大的困扰,我们得想办法把 ...