今天呢在OIwiki上面看到了一个奇怪的生成随机数函数:mt19937年,看到后直接看起来很傻。这是什么?
一开始,我以为这是一个名字M.T.19937937/1937年发明了函数,但回顾过去,他们发现了这一点mt好像是梅森缠绕器(好像有人说是梅森旋转算法),19937是质数,19937是2^19937-1,随后我只是在网上复制了一段代码,粘到DEV里面,然后我震惊地发现代码CE了
然后我抱着尝试的想法复制到最新版本VS2022年,然后我震惊地发现这个东西真的被编译了,生成了一个数字。看完之后,我决定测试这个随机性,所以我有以下代码
#include <cstdio> #include <ctime> #include <iostream> #include <random> using namespace std; int cnt[30050]; struct val { int x;//下标 int y;//数量 }all; int main() { srand(time(0)); std :: mt19937 rng(time(0)); //1-5000,5001-10000,10001-15000,15001-20000,20001-25000,25001-30000; for (int i = 0; i < 1000000; i ) { unsigned int x = rng() % 30000 1; //unsigned int x = rand() % 30000 1; if (1 <= x && x <= 5000) cnt[1] ; else if (5001 <= x && x <= 10000) cnt[2] ; else if (10001 <= x && x <= 15000) cnt[3] ; else if (15001 <= x && x <= 20000) cnt[4] ; else if (20001 <= x && x <= 25000) cnt[5] ; else cnt[6] ; } cout << "第一个数:1-5000 第二个数:5001-10000 第三个数:10001-15000 第四个数:15001-20000 第五个数:20001-25000 第六个数:25001-30000" << endl; cout << cnt[1] << " " << cnt[2] << " " << cnt[3] << " " << cnt[4] << " " << cnt[5] << " " << cnt[6] << endl; memset(cnt, 0, sizeof(cnt)); for (int i = 0; i < 1000000; i ) { unsigned int x = rng() % 30000 1; //unsigned int x = rand() % 30000 1; cnt[x] ; } int maxn = -1, minn = 1000050; for (int i = 1; i <= 30000; i ) { if (cnt[i] > maxn) maxn = cnt[i]; if (cnt[i] < minn) minn = cnt[i]; } cout << "maxn:最多出现在1-30000中,minn:最少出现在1-30000中" << endl; cout << "maxn - minn = " << maxn - minn << endl; long long sum1 = 0, middle, sum2 = 0; for (int i = 1; i <= 30000; i ) { sum1 = i * cnt[i]; if (sum2 < 500000 && sum2 cnt[i] >= 500000) middle = i 1; sum2 = cnt[i]; } double average = (double)sum1 / 1000000.0; cout << "平均数:" << average << endl; cout << "中位数:" << middle << endl; maxn = -1; for (int i = 1; i <= 30000; i ) if (cnt[i] > maxn) maxn = cnt[i]; for (int i = 1; i <= 30000; i ) if (cnt[i] == maxn) all.x = i, all.y = maxn; cout << "众数:" << all.x; return 0; }
经过n次测试,以下是差距最大的组:
可以看到,mt19937的随机性还是很好的,分布均匀,众数随机,平均,中位数接近,maxn-minn相对较小,测试后的随机性,我决定再次测试速度,添加几行代码
#include <cstdio> #include <ctime> #include <iostream> #include <random> using namespace std; int cnt[30050]; struct val { int x;//下标 int y;//数量 }all; int main() { srand(time(0)); std :: mt19937 rng(time(0)); //1-5000,5001-10000,10001-15000,15001-20000,20001-25000,25001-30000; for (int i = 0; i < 1000000; i ) { unsigned int x = rng() % 30000 1; //unsigned int x = rand() % 30000 1; if (1 <= x && x <= 5000) cnt[1] ; else if (5001 <= x && x <= 10000) cnt[2] ; else if (10001 <= x && x <= 15000) cnt[3] ; else if (15001 <= x && x <= 20000) cnt[4] ; else if (20001 <= x && x <= 25000) cnt[5] ; else cnt[6] ; } cout << "第一个数:1-5000 第二个数:5001-10000 第三个数:10001-15000 第四个数:15001-20000 第五个数:20001-25000 第六个数:25001-30000" << endl; cout << cnt[1] << " " << cnt[2] << " " << cnt[3] << " " << cnt[4] << " " << cnt[5] << " " << cnt[6] << endl; memset(cnt, 0, sizeof(cnt)); for (int i = 0; i < 1000000; i ) { unsigned int x = rng() % 30000 1; //unsigned int x = rand() % 30000 1; cnt[x] ; } int maxn = -1, minn = 1000050; for (int i = 1; i <= 30000; i ) { if (cnt[i] > maxn) maxn = cnt[i]; if (cnt[i] < minn) minn = cnt[i]; } cout << "maxn:最多出现在1-30000中,minn:最少出现在1-30000中" << endl; cout << "maxn - minn = " << maxn - minn << endl; long long sum1 = 0, middle, sum2 = 0; for (int i = 1; i <= 30000; i ) { sum1 = i * cnt[i]; if (sum2 < 500000 && sum2 cnt[i] >= 500000) middle = i 1; sum2 = cnt[i]; } double average = (double)sum1 / 1000000.0; cout << "平均数:" << average << endl; cout << "中位数:" << middle << endl; maxn = -1; for (int i = 1; i <= 30000; i ) if (cnt[i] > maxn) maxn = cnt[i]; for (int i = 1; i <= 30000; i ) if (cnt[i] == maxn) all.x = i, all.y = maxn; cout << "众数:" << all.x; clock_t start clock();
for (int i = 0; i < 10000000; i++) {
unsigned int x = rng() % 30000 + 1;
//unsigned int x = rand() % 30000 + 1;
}
cout << endl << "生成10000000个随机数的速度(ms):" << clock() - start;
return 0;
}
这个时候我发现这个mt19937几乎是个O(1)的算法,我准备再测一下rand函数的性能
首先是随机性
代码:
#include <cstdio>
#include <ctime>
#include <iostream>
#include <random>
using namespace std;
int cnt[30050];
struct val {
int x;//下标
int y;//数量
}all;
int main() {
srand(time(0));
std :: mt19937 rng(time(0));
//1-5000,5001-10000,10001-15000,15001-20000,20001-25000,25001-30000;
for (int i = 0; i < 1000000; i++) {
unsigned int x = rng() % 30000 + 1;
//unsigned int x = rand() % 30000 + 1;
if (1 <= x && x <= 5000) cnt[1]++;
else if (5001 <= x && x <= 10000) cnt[2]++;
else if (10001 <= x && x <= 15000) cnt[3]++;
else if (15001 <= x && x <= 20000) cnt[4]++;
else if (20001 <= x && x <= 25000) cnt[5]++;
else cnt[6]++;
}
cout << "第一个数:1-5000 第二个数:5001-10000 第三个数:10001-15000 第四个数:15001-20000 第五个数:20001-25000 第六个数:25001-30000" << endl;
cout << cnt[1] << " " << cnt[2] << " " << cnt[3] << " " << cnt[4] << " " << cnt[5] << " " << cnt[6] << endl;
memset(cnt, 0, sizeof(cnt));
for (int i = 0; i < 1000000; i++) {
unsigned int x = rng() % 30000 + 1;
//unsigned int x = rand() % 30000 + 1;
cnt[x]++;
}
int maxn = -1, minn = 1000050;
for (int i = 1; i <= 30000; i++) {
if (cnt[i] > maxn) maxn = cnt[i];
if (cnt[i] < minn) minn = cnt[i];
}
cout << "maxn:在1-30000中出现最多的数,minn:在1-30000中出现最少的数" << endl;
cout << "maxn - minn = " << maxn - minn << endl;
long long sum1 = 0, middle, sum2 = 0;
for (int i = 1; i <= 30000; i++) {
sum1 += i * cnt[i];
if (sum2 < 500000 && sum2 + cnt[i] >= 500000) middle = i + 1;
sum2 += cnt[i];
}
double average = (double)sum1 / 1000000.0;
cout << "平均数:" << average << endl;
cout << "中位数:" << middle << endl;
maxn = -1;
for (int i = 1; i <= 30000; i++) if (cnt[i] > maxn) maxn = cnt[i];
for (int i = 1; i <= 30000; i++) if (cnt[i] == maxn) all.x = i, all.y = maxn;
cout << "众数:" << all.x;
//clock_t start = clock();
//for (int i = 0; i < 10000000; i++) {
//unsigned int x = rng() % 30000 + 1;
//unsigned int x = rand() % 30000 + 1;
//}
//cout << endl << "生成10000000个随机数的速度(ms):" << clock() - start;
return 0;
}
对比一下和mt19937函数,发现不论是均匀分布还是最大值-最小值,rand都要比mt19937差一大截
最后是rand的时间
代码:
#include <cstdio>
#include <ctime>
#include <iostream>
#include <random>
using namespace std;
int cnt[30050];
struct val {
int x;//下标
int y;//数量
}all;
int main() {
srand(time(0));
std :: mt19937 rng(time(0));
//1-5000,5001-10000,10001-15000,15001-20000,20001-25000,25001-30000;
for (int i = 0; i < 1000000; i++) {
//unsigned int x = rng() % 30000 + 1;
unsigned int x = rand() % 30000 + 1;
if (1 <= x && x <= 5000) cnt[1]++;
else if (5001 <= x && x <= 10000) cnt[2]++;
else if (10001 <= x && x <= 15000) cnt[3]++;
else if (15001 <= x && x <= 20000) cnt[4]++;
else if (20001 <= x && x <= 25000) cnt[5]++;
else cnt[6]++;
}
cout << "第一个数:1-5000 第二个数:5001-10000 第三个数:10001-15000 第四个数:15001-20000 第五个数:20001-25000 第六个数:25001-30000" << endl;
cout << cnt[1] << " " << cnt[2] << " " << cnt[3] << " " << cnt[4] << " " << cnt[5] << " " << cnt[6] << endl;
memset(cnt, 0, sizeof(cnt));
for (int i = 0; i < 1000000; i++) {
//unsigned int x = rng() % 30000 + 1;
unsigned int x = rand() % 30000 + 1;
cnt[x]++;
}
int maxn = -1, minn = 1000050;
for (int i = 1; i <= 30000; i++) {
if (cnt[i] > maxn) maxn = cnt[i];
if (cnt[i] < minn) minn = cnt[i];
}
cout << "maxn:在1-30000中出现最多的数,minn:在1-30000中出现最少的数" << endl;
cout << "maxn - minn = " << maxn - minn << endl;
long long sum1 = 0, middle, sum2 = 0;
for (int i = 1; i <= 30000; i++) {
sum1 += i * cnt[i];
if (sum2 < 500000 && sum2 + cnt[i] >= 500000) middle = i + 1;
sum2 += cnt[i];
}
double average = (double)sum1 / 1000000.0;
cout << "平均数:" << average << endl;
cout << "中位数:" << middle << endl;
maxn = -1;
for (int i = 1; i <= 30000; i++) if (cnt[i] > maxn) maxn = cnt[i];
for (int i = 1; i <= 30000; i++) if (cnt[i] == maxn) all.x = i, all.y = maxn;
cout << "众数:" << all.x;
clock_t start = clock();
for (int i = 0; i < 10000000; i++) {
//unsigned int x = rng() % 30000 + 1;
unsigned int x = rand() % 30000 + 1;
}
cout << endl << "生成10000000个随机数的速度(ms):" << clock() - start;
return 0;
}
然后后我们可以得出1个结论,rand比mt19937除了适用性外都差很多
最后介绍一下用法,就是再代码前面加上这样一行代码初始化
std :: mt19937 rng(time(0));
然后再调用rng()就行了,生成的是unsigned int的范围的随机数,如果用int会有负数
具体的就看OIWIKI吧
最后的最后,大家留个赞再走吧