problems in seeding random number in c++

Question

I used the following code (copied somewhere from the web) and managed to generate some an array of random numbers.

std::default_random_engine generator;
std::uniform_real_distribution<double> distribution(0,1);

int N;
std::cout<< "Please enter the number of disks to be generated: ";
std::cin>> N;

// generates N pairs of coordinates
double** R;
R = new double* [N];
for (int i=0; i<N; i++) {
    R[i] = new double [3];
    for (int j=0; j<3; j++) {
        R[i][j] = distribution(generator);
    }
}

The problem is that the output is always the same, so I think it is not seeded properly. Then I found the following code from the web, but it didn't work on my computer (I always get runtime error)

std::random_device rd;
std::mt19937 gen(rd());
std::uniform_real_distribution<> dis(1, 2);
for (int n = 0; n < 10; ++n) {
    std::cout << dis(gen) << ' ';
}
std::cout << '\n';

What's wrong with the above code? How to seed my code so I can get different outputs everytime? Besides, for the first code, is the linear congruential algorithm being used? How to use 'better' algorithm like the Mersenne Twister (mt19937) ?

thanks

2ns snippet [works for me](https://ideone.com/UYormG). Error is somewhere else — Ivan Aksamentov - Drop, Sep 10 '15 at 10:21
On the C++ side, you can use a `std::vector R(N*3);` to store all those doubles, instead of horribly nested `new[]`. — Tino Didriksen, Sep 10 '15 at 10:23
@Drop it works on gcc, but it can not works on another platforms/compilers. — ForEveR, Sep 10 '15 at 10:26
@Tino Didriksen if I use vector isn't it one-dimensional? I just want to make it two-dimensional so that it 'looks' better to me. — Physicist, Sep 10 '15 at 10:27
2D vector is kind of an abomination when you can use 1D and some simple math, but it's still better than nested new[]. Alternatively, use Boost.Multi_array. — Tino Didriksen, Sep 10 '15 at 10:30
@Drop for example: http://stackoverflow.com/questions/15725677/c0x-random-device-with-stdruntime-error — ForEveR, Sep 10 '15 at 10:32
@ForEveR an isolated ancient compiler/standard library (port) bug doesn't mean the code is incorrect. I mean the answer to a question `"What's wrong with the above code? "` is: "Nothing, it's correct". — Ivan Aksamentov - Drop, Sep 10 '15 at 10:38
@Drop in this case you are right, but `Error is somewhere else` is wrong in my opinion. — ForEveR, Sep 10 '15 at 10:40

ForEveR · Accepted Answer · 2015-09-10T10:40:52.240

4

default_random_engine is implementation-defined. It's not specified which algo it will use.

Which runtime_error you get? It should work correct, but as workaround, you can seed mt19937 with current time as with srand.

std::mt19937 gen(static_cast<std::mt19937::result_type>(std::time(nullptr)));

edited Sep 10 '15 at 10:40

answered Sep 10 '15 at 10:17

ForEveR

55,233
2
119
133

thanks, now it works, but do you mind explaining 'static_cast' in more details please? – Physicist Sep 10 '15 at 10:19
@Physicist `mt19937` ctor receives arg of type `result_type`. `time` will returns `time_t`. So, just cast to `result_type`. – ForEveR Sep 10 '15 at 10:22

score 3 · Answer 2 · answered Sep 10 '15 at 10:16

You need to explicitly seed the random number generator in the constructor with something like this (this example uses the system clock to uniquely seed the generator):

  // obtain a seed from the timer
  unsigned seed = myclock::now().count();
  //construct the generator
  std::default_random_engine generator(seed);

score 0 · Answer 3 · answered Sep 10 '15 at 13:02

You need to provide a random seed for the random number generators. (BTW, the library default of producing the same numbers in the generators is so test results can be predictable). I use the system time as others have shown.

One way I haven't seen mentioned is std::seed_seq which addresses this very issue. Feed is a few integers and it produces a nice sequence of seeds:

#include <random>
#include <vector>
#include <array>
#include <iostream>
#include <iomanip>
#include <chrono>

unsigned long
system_now()
{
  std::chrono::system_clock::time_point now{std::chrono::system_clock::now()};
  std::chrono::system_clock::duration epoch{now.time_since_epoch()};
  return static_cast<unsigned long>(now.time_since_epoch().count());
}

int
main()
{
    // ...
    auto N = 500;

    std::vector<std::array<double, 3>> R(N);

    std::seed_seq seq{1, 2, 3, 4, 5};
    std::vector<std::uint32_t> seeds(1000);
    seq.generate(seeds.begin(), seeds.end());

    std::mt19937 generator(seeds[system_now() % seeds.size()]);
    std::uniform_real_distribution<double> distribution{0.0, 1.0};

    for (auto & V : R)
        for (auto & C : V)
            C = distribution(generator);

    for (const auto & V : R)
    {
        std::cout << '(';
        for (const auto & C : V)
            std::cout << ' ' << std::setw(10) << C;
        std::cout << " )\n";
    }
}

Warning! The seed sequence is also repeatable - with the same input it returns the same output! Also, with 1000 seeds you'll eventually set the same numbers again. Same seed, same generated numbers.

problems in seeding random number in c++

3 Answers3