在做求平面内最近点对的问题时,使用两种不同的比较方法求最小距离,算法效率在100000输入规模下相差2~3倍,很费解,想向大家求助。
具体来说只有求两点距离的方法是不一样的(然而却导致了不同):
一是先一直用平方表示距离,最后输出算开根
if (s == e) return MAX; //如果只有一个点返回无限大
if (s + 1 == e) return square(ar[s], ar[e]);//如果只有两个点返回开根后的距离
二是每次都直接算出开根以后的距离
if (s == e) return MAX; //如果只有一个点返回无限大
if (s + 1 == e) return dis(ar[s], ar[e]);//如果只有两个点返回开根后的距离
以上两种情况都是递归到底后执行,此外还有比较中间是否有最小距离时需要用到
curmin = min(curmin, square(ar[sr[i]], ar[sr[j]]));
curmin = min(curmin, dis(ar[sr[i]], ar[sr[j]]));//(当前求得的两边最小值,中间最小值)
double dis(Node a, Node b) { return sqrt(pow(a.x - b.x, 2) + pow(a.y - b.y, 2)); } //返回点与点之间的距离
double square(Node a, Node b) { return pow(a.x - b.x, 2) + pow(a.y - b.y, 2); } //square()和dis()除了是否开根以外没有任何差别。
两种方法在100000规模下运行20次取平均运行时间如图1和图2所示
图1 dis方法
图2 square方法
输入使用随机数,在随机数中规模大约平均在10的4次方,显然square产生的数字要比dis大得多,每次比较的数字也大的多,比较的复杂度是O(n)的。但是用同样的类型存储数据,也都没有溢出,占位应该是相同的,而且比较是大家都要比较的,dis方法只是数比较小,何况dis每次运算是通过计算平方再开根,多了开根这一步,为什么用dis方法反而会快呢?如何理解数据规模对比较的影响呢?
源码如下:
#include<iostream>
#include<cstdio>
#include<algorithm>
#include<cmath>
#include<time.h>
using namespace std;
double MAX = 1e10; //定义的最大距离,以在只有一个点的时返回无穷大
int a, b; //用来记录下标,与题无关
struct Node {
double x, y;
int key; //关键码,可有可无,与ab有关
};
Node ar[100005];
int sr[100005];
bool cmpx(Node a, Node b) { return a.x<b.x; } //x坐标升序
bool cmpy(Node a, Node b) { return a.y<b.y; } //y坐标升序
int listcmp(const void *a, const void *b)
{
if (ar[*(int*)a].y < ar[*(int*)b].y)//中间的是下标
return -1;
else
return 1;
}
double min(double a, double b) { return a<b ? a : b; } //返回最小值
double dis(Node a, Node b) { return sqrt((a.x - b.x)*(a.x - b.x) + (a.y - b.y)*(a.y - b.y)); } //返回点与点之间的距离
double square(Node a, Node b) {
return (a.x - b.x)*(a.x - b.x) + (a.y - b.y)*(a.y - b.y);
}
void create(int n) {
srand((unsigned)time(NULL));
for (int i = 0; i<n; i++) {
ar[i].key = i + 1;
dr[i].x = rand();
dr[i].y = rand();
//cout << "x " << (ar[i].x = dr[i].x) << " y " << ((ar[i].y = dr[i].y)) << endl;
//cout << "x " << (cr[i].x = dr[i].x) << " y " << ((cr[i].y = dr[i].y)) << endl;
ar[i].x = dr[i].x; ar[i].y = dr[i].y;
cr[i].x = dr[i].x; cr[i].y = dr[i].y;
}
}
double shortest(int s, int e)
{
double d; //d表示点对之间的距离
int listlen = 0;
if (s == e) return MAX; //如果只有一个点
if (s + 1 == e) return dis(ar[s], ar[e]);//如果只有两个点
long i, j, mid = (s + e) >> 1;
double curmin = min(shortest(s, mid), shortest(mid + 1, e));
listlen = 0;
for (i = mid; i >= s && ar[mid + 1].x - ar[i].x <= curmin; i--)
sr[listlen++] = i;
for (i = mid + 1; i <= e && ar[i].x - ar[mid].x <= curmin; i++)
sr[listlen++] = i;
qsort(sr, listlen, sizeof(sr[0]), listcmp);//对y进行排序
for (i = 0; i < listlen; i++)
for (j = i + 1; j < listlen && ar[sr[j]].y - ar[sr[i]].y <= curmin; j++)
curmin = min(curmin, dis(ar[sr[i]], ar[sr[j]]));
return curmin;
}
void myRun (int n) {
time_t start, end;
double distance;
double sum = 0.0;
cout << "分治法在" << n << "规模耗时结果如下:" << endl;
for (int i = 0; i < 20; i++) {
create(n);
sort(ar, ar + n, cmpx); //按x对ar排序
start = clock();
double distance = shortest(0, n);
end = clock();
//if (i == 0) cout << "分治法在"<<n<<"规模时下计算结果如下:"<<endl<<"最短距离为:" << distance << endl<<"时间分别为:";
cout << (double)(end - start) << "ms ";
sum += (double)(end - start);
}
cout <<endl<< "平均耗时: " << sum / 20.0 <<"ms"<< endl<<endl;
}
int main()
{
myRun(100);
myRun(1000);
myRun(10000);
myRun(20000);
myRun(40000);
myRun(60000);
myRun(80000);
myRun(100000);
system("pause");
return 0;
}
Don’t know the rest of you.
In many cases, I don’t prescribe a prescription to judge the size or other things. I only prescribe a prescription when the result is needed.
This saves you one prescription when judging the size, which can save a lot.
Personally, I think it is most likely that other parts of you caused this result.