Description

You are working for an amusement park as an operator of an obakeyashiki, or a haunted house, in which guests walk through narrow and dark corridors. The house is proud of their lively ghosts, which are actually robots remotely controlled by the operator, hiding here and there in the corridors. One morning, you found that the ghosts are not in the positions where they are supposed to be. Ah, yesterday was Halloween. Believe or not, paranormal spirits have moved them around the corridors in the night. You have to move them into their right positions before guests come. Your manager is eager to know how long it takes to restore the ghosts.

In this problem, you are asked to write a program that, given a floor map of a house, finds the smallest number of steps to move all ghosts to the positions where they are supposed to be.

A floor consists of a matrix of square cells. A cell is either a wall cell where ghosts cannot move into or a corridor cell where they can.

At each step, you can move any number of ghosts simultaneously. Every ghost can either stay in the current cell, or move to one of the corridor cells in its 4-neighborhood (i.e. immediately left, right, up or down), if the ghosts satisfy the following conditions:

  1. No more than one ghost occupies one position at the end of the step.

  2. No pair of ghosts exchange their positions one another in the step.

For example, suppose ghosts are located as shown in the following (partial) map, where a sharp sign (‘#’) represents a wall cell and ‘a’, ‘b’, and ‘c’ ghosts.

####
ab#
#c##
####

The following four maps show the only possible positions of the ghosts after one step.

####
ab#
#c##
####
 
####
a b#
#c##
####
 
####
acb#
# ##
####
 
####
ab #
#c##
####

Input

The input consists of at most 10 datasets, each of which represents a floor map of a house. The format of a dataset is as follows.

w h n  
c11 c12 ⋯ c1w
c21 c22 ⋯ c2w
⋮ ⋮ ⋱ ⋮
ch1 ch2 ⋯ chw

w, h and n in the first line are integers, separated by a space. w and h are the floor width and height of the house, respectively. n is the number of ghosts. They satisfy the following constraints.

4 ≤ w ≤ 16, 4 ≤ h ≤ 16, 1 ≤ n ≤ 3

Subsequent h lines of w characters are the floor map. Each of cij is either:

  • a ‘#’ representing a wall cell,

  • a lowercase letter representing a corridor cell which is the initial position of a ghost,

  • an uppercase letter representing a corridor cell which is the position where the ghost corresponding to its lowercase letter is supposed to be, or

  • a space representing a corridor cell that is none of the above.

In each map, each of the first n letters from a and the first n letters from A appears once and only once. Outermost cells of a map are walls; i.e. all characters of the first and last lines are sharps; and the first and last characters on each line are also sharps. All corridor cells in a map are connected; i.e. given a corridor cell, you can reach any other corridor cell by following corridor cells in the 4-neighborhoods. Similarly, all wall cells are connected. Any 2 × 2 area on any map has at least one sharp. You can assume that every map has a sequence of moves of ghosts that restores all ghosts to the positions where they are supposed to be.

The last dataset is followed by a line containing three zeros separated by a space.

Output

For each dataset in the input, one line containing the smallest number of steps to restore ghosts into the positions where they are supposed to be should be output. An output line should not contain extra characters such as spaces.

Sample Input

5 5 2
#####
#A#B#
# #
#b#a#
#####
16 4 3
################
## ########## ##
# ABCcba #
################
16 16 3
################
### ## # ##
## # ## # c#
# ## ########b#
# ## # # # #
# # ## # # ##
## a# # # # #
### ## #### ## #
## # # # #
# ##### # ## ##
#### #B# # #
## C# # ###
# # # ####### #
# ###### A## #
# # ##
################
0 0 0

Sample Output

7
36
77


这题直接搜索就可过,不用双向搜索。
但双向搜索比直接搜索快了一倍...
因为题目中说明障碍很多,如果我们对一个状态向四面扩展判断是否可行的话,就会多做非常多的运算和判断,常数巨大。
我们可以处理出每个点向四周有哪些可以走,这样会大大优化常数。
bfs要判重需要开很大的数组,在poj上开17*17的3次方的数组会炸掉。
所以一个障碍物不给他标号,因为他们不可能被用到,这样就少了很多无用的空间。
为了让代码更加简便,处理3只鬼以下的情况,我们只用在图外面新开一个点,当做缺少的鬼的起点和终点,避免的过多的分类讨论。
#include <iostream>
#include <cstdio>
#include <cstring>
#include <queue>
#include <vector>
#include <cstdlib>
using namespace std;
#define reg register
const int dx[] = {, , -, , }, dy[] = {, , , , -};
int n, m, Gh;
char mp[][];
int S[], T[];
int id[][], cnt, X[*], Y[*];
bool vis[*][*][*];
vector <int> ve[*]; struct date {
int a, b, c;
int stp;
}; bool ok(int x1, int y1, int x2, int y2) {
if (x2 == y2) return ;
if (x1 == y2 and y1 == x2) return ;
return ;
} int main()
{
while()
{
for (reg int i = ; i <= cnt ; i ++) ve[i].clear();
cnt = ;
scanf("%d%d%d", &m, &n, &Gh);
if (!n and !m and !Gh) break;
char c;
while((c = getchar()) || ) if(c == '\n') break;
for (reg int i = ; i <= n ; i ++)
{
gets(mp[i]);
for (reg int j = ; j < m ; j ++)
{
id[i][j + ] = ++cnt;
X[cnt] = i, Y[cnt] = j + ;
if (mp[i][j] >= 'a' and mp[i][j] <= 'c') S[mp[i][j] - 'a' + ] = id[i][j + ];
if (mp[i][j] >= 'A' and mp[i][j] <= 'C') T[mp[i][j] - 'A' + ] = id[i][j + ];
}
}
for (reg int i = ; i <= n ; i ++)
{
for (reg int j = ; j <= m ; j ++)
{
for (reg int k = ; k <= ; k ++)
{
int ti = i + dx[k], tj = j + dy[k];
if (ti <= or ti > n or tj <= or tj > m or mp[ti][tj - ] == '#') continue;
ve[id[i][j]].push_back(id[ti][tj]);
}
}
}
if (Gh == ) {
S[] = cnt + , T[] = cnt + ;
S[] = cnt + , T[] = cnt + ;
}
if (Gh == ) {
S[] = cnt + , T[] = cnt + ;
}
ve[cnt + ].push_back(cnt + );
ve[cnt + ].push_back(cnt + );
queue <date> q;
q.push((date){S[], S[], S[], });
memset(vis, , sizeof vis);
vis[S[]][S[]][S[]] = ; while(!q.empty())
{
date ft = q.front();q.pop();
if (ft.a == T[] and ft.b == T[] and ft.c == T[]) {printf("%d\n", ft.stp);goto End;}
for (reg int i = ; i < (signed)ve[ft.a].size() ; i ++)
{
for (reg int j = ; j < (signed)ve[ft.b].size() ; j ++)
{
if (!ok(ft.a, ft.b, ve[ft.a][i], ve[ft.b][j])) continue;
for (reg int k = ; k < (signed)ve[ft.c].size() ; k ++)
{
if (vis[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]]) continue;
if (!ok(ft.a, ft.c, ve[ft.a][i], ve[ft.c][k]) or !ok(ft.b, ft.c, ve[ft.b][j], ve[ft.c][k])) continue;
vis[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]] = ;
q.push(((date){ve[ft.a][i], ve[ft.b][j], ve[ft.c][k], ft.stp + }));
}
}
}
}
End:;
}
}

暴力搜索

双向搜索直接在以上代码上改改就行了。开两个队列,分别从初始状态和末尾状态同时往里搜索,每次各扩展一层,如果相遇则得到答案。

 
#include <iostream>
#include <cstdio>
#include <cstring>
#include <queue>
#include <vector>
#include <cstdlib>
#include <string>
using namespace std;
#define reg register
const int dx[] = {, , -, , }, dy[] = {, , , , -};
int n, m, Gh;
string mp[];
int S[], T[];
int id[][], cnt, X[], Y[];
int vis1[][][], vis2[][][];
vector <int> ve[]; struct date {
int a, b, c;
int stp;
}; bool ok(int x1, int y1, int x2, int y2) {
if (x2 == y2) return ;
if (x1 == y2 and y1 == x2) return ;
return ;
} int main()
{
while()
{
for (reg int i = ; i <= cnt ; i ++) ve[i].clear();
cnt = ;
scanf("%d%d%d", &m, &n, &Gh);
if (!n and !m and !Gh) break;
char c;
while((c = getchar()) || ) if(c == '\n') break;
for (reg int i = ; i <= n ; i ++)
{
getline(cin, mp[i]);
for (reg int j = ; j < m ; j ++)
{
if (mp[i][j] != '#')
id[i][j + ] = ++cnt, X[cnt] = i, Y[cnt] = j + ;
if (mp[i][j] >= 'a' and mp[i][j] <= 'c') S[mp[i][j] - 'a' + ] = id[i][j + ];
if (mp[i][j] >= 'A' and mp[i][j] <= 'C') T[mp[i][j] - 'A' + ] = id[i][j + ];
}
}
for (reg int i = ; i <= n ; i ++)
{
for (reg int j = ; j <= m ; j ++)
{
if (mp[i][j - ] == '#') continue;
for (reg int k = ; k <= ; k ++)
{
int ti = i + dx[k], tj = j + dy[k];
if (ti <= or ti > n or tj <= or tj > m or mp[ti][tj - ] == '#') continue;
ve[id[i][j]].push_back(id[ti][tj]);
}
}
}
if (Gh == ) {
S[] = cnt + , T[] = cnt + ;
S[] = cnt + , T[] = cnt + ;
}
if (Gh == ) {
S[] = cnt + , T[] = cnt + ;
}
ve[cnt + ].push_back(cnt + );
ve[cnt + ].push_back(cnt + ); queue <date> q1, q2;
q1.push((date){S[], S[], S[], });
q2.push((date){T[], T[], T[], });
memset(vis1, , sizeof vis1);
memset(vis2, , sizeof vis2);
vis1[S[]][S[]][S[]] = ;
vis2[T[]][T[]][T[]] = ;
while(!q1.empty() and !q2.empty())
{
date ft = q1.front();q1.pop();
if (vis2[ft.a][ft.b][ft.c]) {printf("%d\n", ft.stp + vis2[ft.a][ft.b][ft.c]);goto End;};
for (reg int i = ; i < (signed)ve[ft.a].size() ; i ++)
{
for (reg int j = ; j < (signed)ve[ft.b].size() ; j ++)
{
if (!ok(ft.a, ft.b, ve[ft.a][i], ve[ft.b][j])) continue;
for (reg int k = ; k < (signed)ve[ft.c].size() ; k ++)
{
if (vis1[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]]) continue;
if (!ok(ft.a, ft.c, ve[ft.a][i], ve[ft.c][k]) or !ok(ft.b, ft.c, ve[ft.b][j], ve[ft.c][k])) continue;
vis1[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]] = ft.stp + ;
q1.push(((date){ve[ft.a][i], ve[ft.b][j], ve[ft.c][k], ft.stp + }));
}
}
} ft = q2.front();q2.pop();
if (vis1[ft.a][ft.b][ft.c]) {printf("%d\n", ft.stp + vis1[ft.a][ft.b][ft.c]);goto End;};
for (reg int i = ; i < (signed)ve[ft.a].size() ; i ++)
{
for (reg int j = ; j < (signed)ve[ft.b].size() ; j ++)
{
if (!ok(ft.a, ft.b, ve[ft.a][i], ve[ft.b][j])) continue;
for (reg int k = ; k < (signed)ve[ft.c].size() ; k ++)
{
if (vis2[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]]) continue;
if (!ok(ft.a, ft.c, ve[ft.a][i], ve[ft.c][k]) or !ok(ft.b, ft.c, ve[ft.b][j], ve[ft.c][k])) continue;
vis2[ve[ft.a][i]][ve[ft.b][j]][ve[ft.c][k]] = ft.stp + ;
q2.push(((date){ve[ft.a][i], ve[ft.b][j], ve[ft.c][k], ft.stp + }));
}
}
} }
End:;
}
return ;
}

 

[POJ3523]The Morning after Halloween的更多相关文章

  1. 【BFS】The Morning after Halloween

    [POJ3523]The Morning after Halloween Time Limit: 8000MS   Memory Limit: 65536K Total Submissions: 23 ...

  2. POJ 3370. Halloween treats 抽屉原理 / 鸽巢原理

    Halloween treats Time Limit: 2000MS   Memory Limit: 65536K Total Submissions: 7644   Accepted: 2798 ...

  3. Lightoj 题目1422 - Halloween Costumes(区间DP)

    1422 - Halloween Costumes   PDF (English) Statistics Forum Time Limit: 2 second(s) Memory Limit: 32 ...

  4. CSUFT 1004 This is Halloween: Saving Money

    1004: This is Halloween: Saving Money Time Limit: 1 Sec      Memory Limit: 128 MB Submit: 11      So ...

  5. [POJ 3370] Halloween treats

    Halloween treats Time Limit: 2000MS   Memory Limit: 65536K Total Submissions: 7143   Accepted: 2641 ...

  6. poj 3370 Halloween treats(鸽巢原理)

    Description Every year there is the same problem at Halloween: Each neighbour is only willing to giv ...

  7. LightOJ - 1422 Halloween Costumes (区间dp)

    Description Gappu has a very busy weekend ahead of him. Because, next weekend is Halloween, and he i ...

  8. UVA 11237 - Halloween treats(鸽笼原理)

    11237 - Halloween treats option=com_onlinejudge&Itemid=8&page=show_problem&category=516& ...

  9. 鸽巢原理应用-分糖果 POJ 3370 Halloween treats

    基本原理:n+1只鸽子飞回n个鸽笼至少有一个鸽笼含有不少于2只的鸽子. 很简单,应用却也很多,很巧妙,看例题: Description Every year there is the same pro ...

随机推荐

  1. OPC质量列表(OPC quality of the list)质量戳定义

    参考:<OPC统一架构>中文版 0 Bad1 Bad, Low Limited2 Bad, High Limited3 Bad, Constant4 Bad, Configuration ...

  2. jvm 内存溢出

    堆内存溢出 堆内存中存在大量对象,这些对象都有被引用,当所有对象占用空间达到堆内存的最大值,就会出现内存溢出OutOfMemory:Java heap space 永久代溢出 类的一些信息,如类名.访 ...

  3. Tomcat9 安装与配置

    一.下载 到http://tomcat.apache.org/下载绿色解压包 二.启动 1.解压后打开tomcat/bin目录下的startup.bat即可启动 打开后发现出现乱码 解决方法: 打开t ...

  4. MySQL设计表规范

    规范总结 所有数据库对象名称必须使用小写字母并用下划线分割 所有数据库对象名称禁止使用 MySQL 保留关键字[设计表后逐一排查] 所有表必须使用 Innodb 存储引擎,数据库和表的字符集统一使用 ...

  5. [CUDA] 00 - GPU Driver Installation & Concurrency Programming

    前言 对,这是一个高大上的技术,终于要做老崔当年做过的事情了,生活很传奇. 一.主流 GPU 编程接口 1. CUDA 是英伟达公司推出的,专门针对 N 卡进行 GPU 编程的接口.文档资料很齐全,几 ...

  6. 从零开始OpenGL—— 一、 环境配置

    前言 高考完之后填志愿,当时想以后去做游戏,所以选择了计算机这个专业,之前捣鼓过U3D,这学期也开始了计算机图形学的学习,最近学习了OpenGL相关的一些内容,将在博客中记录这系列的学习.这篇开篇博客 ...

  7. Pytorch-数学运算

    引言 本篇介绍tensor的数学运算. 基本运算 add/minus/multiply/divide matmul pow sqrt/rsqrt round 基础运算 可以使用 + - * / 推荐 ...

  8. hadoop之mapreduce详解(进阶篇)

    上篇文章hadoop之mapreduce详解(基础篇)我们了解了mapreduce的执行过程和shuffle过程,本篇文章主要从mapreduce的组件和输入输出方面进行阐述. 一.mapreduce ...

  9. Spring框架学习笔记(3)——SpringMVC框架

    SpringMVC框架是基于Spring框架,可以让我们更为方便的进行Web的开发,实现前后端分离 思路和原理 我们之前仿照SpringMVC定义了一个自定义MVC框架,两者的思路其实都是一样的. 建 ...

  10. 注意!GetThreadPriority的返回值不是系统的优先级值

    GetThreadPriority的返回值 Return code/value Description THREAD_PRIORITY_ABOVE_NORMAL 1 Priority 1 point ...