1//===----------------------------------------------------------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is dual licensed under the MIT and the University of Illinois Open
6// Source Licenses. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9
10// <random>
11
12// class bernoulli_distribution
13
14// template<class _URNG> result_type operator()(_URNG& g);
15
16#include <random>
17#include <numeric>
18#include <vector>
19#include <cassert>
20
21template <class T>
22inline
23T
24sqr(T x)
25{
26    return x * x;
27}
28
29int main()
30{
31    {
32        typedef std::bernoulli_distribution D;
33        typedef std::minstd_rand G;
34        G g;
35        D d(.75);
36        const int N = 100000;
37        std::vector<D::result_type> u;
38        for (int i = 0; i < N; ++i)
39            u.push_back(d(g));
40        double mean = std::accumulate(u.begin(), u.end(),
41                                              double(0)) / u.size();
42        double var = 0;
43        double skew = 0;
44        double kurtosis = 0;
45        for (int i = 0; i < u.size(); ++i)
46        {
47            double d = (u[i] - mean);
48            double d2 = sqr(d);
49            var += d2;
50            skew += d * d2;
51            kurtosis += d2 * d2;
52        }
53        var /= u.size();
54        double dev = std::sqrt(var);
55        skew /= u.size() * dev * var;
56        kurtosis /= u.size() * var * var;
57        kurtosis -= 3;
58        double x_mean = d.p();
59        double x_var = d.p()*(1-d.p());
60        double x_skew = (1 - 2 * d.p())/std::sqrt(x_var);
61        double x_kurtosis = (6 * sqr(d.p()) - 6 * d.p() + 1)/x_var;
62        assert(std::abs((mean - x_mean) / x_mean) < 0.01);
63        assert(std::abs((var - x_var) / x_var) < 0.01);
64        assert(std::abs((skew - x_skew) / x_skew) < 0.01);
65        assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.02);
66    }
67    {
68        typedef std::bernoulli_distribution D;
69        typedef std::minstd_rand G;
70        G g;
71        D d(.25);
72        const int N = 100000;
73        std::vector<D::result_type> u;
74        for (int i = 0; i < N; ++i)
75            u.push_back(d(g));
76        double mean = std::accumulate(u.begin(), u.end(),
77                                              double(0)) / u.size();
78        double var = 0;
79        double skew = 0;
80        double kurtosis = 0;
81        for (int i = 0; i < u.size(); ++i)
82        {
83            double d = (u[i] - mean);
84            double d2 = sqr(d);
85            var += d2;
86            skew += d * d2;
87            kurtosis += d2 * d2;
88        }
89        var /= u.size();
90        double dev = std::sqrt(var);
91        skew /= u.size() * dev * var;
92        kurtosis /= u.size() * var * var;
93        kurtosis -= 3;
94        double x_mean = d.p();
95        double x_var = d.p()*(1-d.p());
96        double x_skew = (1 - 2 * d.p())/std::sqrt(x_var);
97        double x_kurtosis = (6 * sqr(d.p()) - 6 * d.p() + 1)/x_var;
98        assert(std::abs((mean - x_mean) / x_mean) < 0.01);
99        assert(std::abs((var - x_var) / x_var) < 0.01);
100        assert(std::abs((skew - x_skew) / x_skew) < 0.01);
101        assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.02);
102    }
103}
104