Time Limit: 20000/10000MS (Java/Others) Memory Limit: 128000/64000KB (Java/Others)     Special Judge

Problem Description

Most modern archivers, such as WinRAR or WinZIP, use modifications of Lempel-Ziv method as their primary compression algorithm. Although decompression of LZ-compressed archives is usually easy and fast, the compression process itself is often rather complicated and slow. Therefore professional archivers use approximation methods that sometimes do not allow to achieve the best possible compression.

This situation doesn’t satisfy your chief George. He would like to create the best archiver WinGOR. The archiver will use the following modification of LZ77 algorithm.
      The text is partitioned to chunks of length not exceeding 4096. Each chunk is compressed independently. We will describe the decompression of one chunk t. Based on this description, you will have to create a compression algorithm that will create the shortest possible compressed chunk x from the given chunk t.
      The compressed chunk is written down as the sequence of plain characters and repetition blocks. Plain character is 8 bits long. When decompressing, plain character c is simply copied to output. Repetition block (r; l) consists of two parts: reference r and length l, each 12 bits long. Reference r is an integer number between 1 and 4095. When repetition block (r; l) is obtained after decompressing i − 1 characters of text, characters t[i − r ... i − r + l − 1] are copied to output. Note, that r can be less than l, in this
case recently copied characters are copied to output as well.

To help decompressor distinguish between plain characters and repetition blocks a leading bit is prepended to each element of the compressed text: 0 means plain character follows, 1 — repetition block follows.

For example, “aaabbaaabababababab” can be compressed as “aaabb(5,4)(2,10)”. The compressed variant has 8 + 8 + 8 + 8 + 8 + 24 + 24 + 7 = 95 bits instead of 152 in the original text (additional 7 bits are used to distinguish between plain characters and repetition blocks).
      Given a text chunk, find its compressed representation which needs fewest number of bits to encode

Input

      Input file contains a text chunk t. Its length doesn’t exceed 4096. A text chunk contains only small letters of the English alphabet.

Output

      Print the length of the compressed text in bits at the first line of the output file. Print the compressed chunk itself at the second line of the output file. Use characters themselves to denote plain characters and “(r,l)” notation (without spaces) to denote repetition blocks.

Sample Input

aaabbaaabababababab 

Sample Output

95 
aaabb(5,4)(2,10)

题意概述

给定一种压缩规则:或直接复制,或将重复出现的字符用二元数对(r,l)表示,代表[i − r ... i − r + l − 1]范围内的字符串。其中直接复制的字符占9位,每个二元数对占25位。对于给定的待压缩字符串,要求输出压缩后的最小数据位数和压缩后的编码。

分析

容易看出,字符串中前i位如何编码,对后继字符串的编码是没有影响的。这样,我们就可以想到运用保存决策的动态规划进行编码。
这里需要的一点小trick在于如何寻找尽量靠前的一段最长重复子串。代入kmp算法的思路,我们可以对原字符串的每一段后缀预处理出next数组,这样原字符串中每个前缀的最长重复子串就可以在O(n)的时间内得到了。于是这里可以得出一个复杂度O(n^2)的动态规划:
记minbit[i]为压缩前i个字符所需的最小数据量,str[i]为此时的决策,r[i],l[i]为这次转移时压缩成的数对(若不需压缩,str[i]=i, r[i]储存不需压缩的部分的起始点)
那么,每次转移时就可以从0到i-1枚举决策j,选择可以使(minbit[j] + (i-j)*9)或minbit[i - next[j][i-1]] + 25取得最小值的j,并保存决策。最后可以通过一个递归函数推出最终答案。
(这里需要注意一点问题……next数组是二维的,数据范围maxn为2^12,而内存限制为64000KB……如果用next[maxn][maxn]的方式开数组明显会MLE……所以我鼓起勇气采用了动态内存分配……)

AC代码

 1 //Verdict: Accepted 2 // Submission Date: -- ::
 3 // Time: 8256MS
 4 // Memory: 34624KB
 5
 6 /*=============================================================================================================================*/
 7 /*======================================================Code by Asm.Def========================================================*/
 8 /*=============================================================================================================================*/
 9 #include <cstdio>
 #include <iostream>
 #include <algorithm>
 #include <cmath>
 #include <cctype>
 #include <memory.h>
 #include <cstring>
 #include <cstdlib>
 using namespace std;
 #define maxn ((int)4.1e3)
 /*===========================================================TYPES=============================================================*/
 typedef long long LL;
   
 /*======================================================GLOBAL VARIABLES=======================================================*/
 char ch[maxn];
 int len = , minbit[maxn], *next[maxn];
 int l[maxn], r[maxn], str[maxn];
 /*==========================================================FUNCTIONS==========================================================*/
 inline void getnext(int l){
     int i, j, L = len - l;
     next[l] = new int[L+];
     int *Next = next[l];
     Next[] = ;
     for(i = ;i < L;++i){
         j = Next[i-] - ;
         while(ch[l+i] != ch[l+j+] && j >= )
             j = Next[j] - ;
         if(ch[l+i] == ch[l+j+])
             Next[i] = j + ;
         else Next[i] = ;
     }
 }
 void printpro(int i){
     if(str[i] == i){
         if(r[i])printpro(r[i]-);
         int j;
         for(j = r[i];j <= i;++j)putchar(ch[j]);
         return;
     }
     printpro(str[i]);
     printf("(%d,%d)", r[i], l[i]);
 }
 int main(){
     #ifdef DEBUG
     assert(freopen("test","r",stdin));
     #endif
     //--------------------------------------------------variables-----------------------------------------------------------
       
     //-----------------------------------------------------work-------------------------------------------------------------
     char c;
     while(isalpha(c = getchar()))str[len] = len, ch[len++] = c;
     int i, j, Min, t;
     for(i = ;i < len - ; ++i)
         getnext(i);
     minbit[] = ;
     for(i = ;i < len; ++i){
         Min = 0x7fffffff;
         for(j = ;j < i;++j)
             if(minbit[j] + (i-j)* < Min){
                 Min = minbit[j] + (i-j)*;
                 str[i] = i;
                 r[i] = j+;
             }
         for(j = ;j < i; ++j){
             t = next[j][i-j];
             if(!t)continue;
             if(minbit[i-t] +  < Min){
                 Min = minbit[i-t] + ;
                 str[i] = i-t;
                 r[i] = i+-t-j;
                 l[i] = t;
             }
         }
         minbit[i] = Min;
     }
     printf("%d\n", minbit[len-]);
     printpro(len-);
     return ;
 }
 /*=============================================================================================================================*/ 

[Andrew Stankevich's Contest#21] Lempel-Ziv Compression的更多相关文章

  1. Andrew Stankevich's Contest (21) J dp+组合数

    坑爹的,,组合数模板,,, 6132 njczy2010 1412 Accepted 5572 MS 50620 KB C++ 1844 B 2014-10-02 21:41:15 J - 2-3 T ...

  2. 【模拟ACM排名】ZOJ-2593 Ranking (Andrew Stankevich’s Contest #5)

    真心是道水题,但找bug找的我想剁手了/(ㄒoㄒ)/~~ 注意几个坑点, 1.输入,getline(cin); / gets(); 一行输入,注意前面要加getchar();   输入运行记录的时候可 ...

  3. Andrew Stankevich&#39;s Contest (1)

    Andrew Stankevich's Contest (1) 打一半出门了,回来才补完了...各种大数又不能上java..也是蛋疼无比 A:依据置换循环节非常easy得出要gcd(x, n) = 1 ...

  4. acdream:Andrew Stankevich Contest 3:Two Cylinders:数值积分

    Two Cylinders Special JudgeTime Limit: 10000/5000MS (Java/Others)Memory Limit: 128000/64000KB (Java/ ...

  5. GYM 100608G 记忆化搜索+概率 2014-2015 Winter Petrozavodsk Camp, Andrew Stankevich Contest 47 (ASC 47)

    https://codeforces.com/gym/100608 题意: 两个人玩游戏,每个人有一个长为d的b进制数字,两个人轮流摇一个$[0,b-1]$的骰子,并将选出的数字填入自己的d个空位之中 ...

  6. LeetCode Weekly Contest 21

    1. 530. Minimum Absolute Difference in BST 最小的差一定发生在有序数组的相邻两个数之间,所以对每一个数,找他的前驱和后继,更新结果即可!再仔细一想,bst的中 ...

  7. NOIP模拟·20141105题解

    [A.韩信点兵] 结论题+模板题,用到了中国剩余定理,维基百科上讲的就比较详细,这里就不再赘述了…… 对于这题,我们先利用中国剩余定理($x \equiv \sum{(a_i m_i (m_i^{-1 ...

  8. ZJU 2605 Under Control

    Under Control Time Limit: 2000ms Memory Limit: 65536KB This problem will be judged on ZJU. Original ...

  9. AC自动机-算法详解

    What's Aho-Corasick automaton? 一种多模式串匹配算法,该算法在1975年产生于贝尔实验室,是著名的多模式匹配算法之一. 简单的说,KMP用来在一篇文章中匹配一个模式串:但 ...

随机推荐

  1. python爬虫模块之URL管理器模块

    URL管理器模块 一般是用来维护爬取的url和未爬取的url已经新添加的url的,如果队列中已经存在了当前爬取的url了就不需要再重复爬取了,另外防止造成一个死循环.举个例子 我爬www.baidu. ...

  2. 2015多校第6场 HDU 5354 Bipartite Graph CDQ,并查集

    题目链接:http://acm.hdu.edu.cn/showproblem.php?pid=5354 题意:求删去每个点后图是否存在奇环(n,m<=1e5) 解法:很经典的套路,和这题一样:h ...

  3. C后端设计开发 - 第3章-气功-原子锁线程协程

    正文 第3章-气功-原子锁线程协程 后记 如果有错误, 欢迎指正. 有好的补充, 和疑问欢迎交流, 一块提高. 在此谢谢大家了. 童话镇 - http://music.163.com/#/m/song ...

  4. ASPxgridview 编辑列初始化事件

    在初始化编辑咧的时候,给其赋值或者是disable等等.... 贴上代码 protected void master_CellEditorInitialize(object sender, ASPxG ...

  5. Leetcode 之Balanced Binary Tree(49)

    用递归的方式来做,左右两棵子树的高度差不超过1.分成两部分,一部分递归得到树的高度,一部分递归检查左右子树是否是平衡二叉树. int getHeight(TreeNode *root) { ; ; } ...

  6. 《深入浅出MyBatis技术原理与实战》——4. 映射器,5. 动态SQL

    4.1 映射器的主要元素 4.2 select元素 4.2.2 简易数据类型的例子 例如,我们需要统计一个姓氏的用户数量.应该把姓氏作为参数传递,而将结果设置为整型返回给调用者,如: 4.2.3 自动 ...

  7. hdu 1226(同余搜索)

    超级密码 Time Limit: 20000/10000 MS (Java/Others)    Memory Limit: 65536/32768 K (Java/Others)Total Subm ...

  8. AC日记——「SCOI2015」情报传递 LiBreOJ 2011

    #2011. 「SCOI2015」情报传递 思路: 可持久化树状数组模板: 代码: #include <bits/stdc++.h> using namespace std; #defin ...

  9. AC日记——贪婪大陆 洛谷 P2184

    贪婪大陆 思路: 树状数组: 跪烂.. 代码: #include <bits/stdc++.h> using namespace std; #define maxn 100005 int ...

  10. windows下rabbitmq(架构师必备神器)集群搭建

    准备2台机器,例如:computera: 10.0.0.151   computerb:10.0.0.234  都安装erlang环境和rabbitmq服务,注意otp环境和rabbitmq服务必须版 ...