Generating a multinomial distribution

Question

I am trying to write/find a code piece/function which gives back a multinomial distribution given a number and a probability array, say a function:

Mult(N, pp[])

For example, for pp={0.3, 0.2, 0.5}, (pp can have variable sizes, say 2 to 10 elements).

Mult(10, pp[])

giving outputs like:

1,1,8
3,1,6
2,3,5

when called in different occasions.

I can think of something like the following using gsl but when called repeatedly in different occasions this is not good. I understand that the code piece I quoted below is not a good one, but looking forward for suggestions to have a better one. Thanks.

std::vector<int> Mult(int numb, std::vector<double> prob_array)
{
    const gsl_rng_type * T2;
    gsl_rng * r2;
    srand(); // srand(time(NULL));

    unsigned int Seed2 = 1234567; // rand();
    gsl_rng_env_setup();

    T2 = gsl_rng_default;
    r2 = gsl_rng_alloc (T2);
    gsl_rng_set (r2, Seed2);

    size_t k = prob_array.size();

    double ppp[k]; // Probability array

    for(int ii=0; ii<prob_array.size(); ++ii) {
        ppp[ii] =  prob_array[ii];
    }

    unsigned int mult_op[k];
    gsl_ran_multinomial(r2, k, numb, ppp, mult_op);
    std::vector<int> multi;
    for(int ii=0; ii<kk; ++ii ){
        multi.push_back(mult_op[ii]);
    }
    return multi;
}

Have you considered using C++11 std::binomial_distribution instead of gsl? — DoctorMoisha, Aug 18 '14 at 13:27
@DoctorMoisha, for p={p,(1.0-p)} cases, yes, I am using std::binomial_distribution. Is std::binomial_distribution extendable to multinomial? if yes, please elaborate. — Attu nonUser De, Aug 18 '14 at 13:53
Sorry, my bad. No it is not extendable. Maybe this link will help https://gist.github.com/miki16g/8319640 — DoctorMoisha, Aug 18 '14 at 14:04
Thanks @DoctorMoisha, I was also looking at the same source. :) — Attu nonUser De, Aug 18 '14 at 14:07
repeated application of [std::discrete_distribution](http://en.cppreference.com/w/cpp/numeric/random/discrete_distribution) might be useful - it appears to be multinomial with one trial — Aaron McDaid, Jan 07 '17 at 12:41

Generating a multinomial distribution

0 Answers0