Thread: Ranom numbers having uniform distribution

If we assume that "uniform" means "it's roughly the same likelihood of getting all numbers in the expected range", then I'd agree with Tabstop. rand() is not perfect, but for many things it's good enough. And this sounds like one of those cases.

If you intend to make money using the random numbers, then you may need something better (less predictable for example), so that someone playing your card game or whatever can not predict what the next set of cards will be, for example.

--
Mats

Originally Posted by edesign

I want it to be uniformly distributed precisely. If I generate a histogram, I want it to be flat, same count for all the numbers. I need to generate the numbers in [0,1]. If rand generates same number more times than it'll destroy Uniform Dist property and will lead to errors in further calculations. The only way I can think of avoiding it is to check for every value, but that will be too costly to do. Is there some other way to do that?

In other words, when you generate say the 400th random number it is absolutely not allowed under any circumstance to be equal to one of the previous 399 numbers?
If so, then you can't just use a PRNG directly. You need an algorithm. The good news is that when done cleverly the cost of checking that the number hasn't been generated already becomes near zero.
The trick is to first set up an array of the items you want to include in your output, pick one at random, and then overwrite it with one from point n, decreasing the length of the portion you are allowed to pick numbers from.
My C is a little rusty, I'm used to C++ but...

Code:

int main(void)
{
    /* setup */
    float values[500];
    int n = 500;
    int pos;
    for (int i=0; i<500; ++i)
        values[i] = i / 499.0;

    /* generation */
    srand(time(NULL));
    while (n > 0)
    {
        pos = rand() % n;
        printf("%f\n", values[pos]);
        n--;
        values[pos] = values[n];
    }
}

Including the appropriate headers is left as an exercise fo the reader
That should give 500 equally spaces values in the range [0,1] assuming you didn't actually mean the range [0,1).

Originally Posted by iMalc

You need an algorithm. The good news is that when done cleverly the cost of checking that the number hasn't been generated already becomes near zero.
The trick is to first set up an array of the items you want to include in your output, pick one at random, and then overwrite it with one from point n, decreasing the length of the portion you are allowed to pick numbers from.

That's exactly what I suggested, using a Durstenfeld style shuffle.