• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 // template<class IntType = int>
13 // class negative_binomial_distribution
14 
15 // template<class _URNG> result_type operator()(_URNG& g);
16 
17 #include <random>
18 #include <numeric>
19 #include <vector>
20 #include <cassert>
21 
22 template <class T>
23 inline
24 T
sqr(T x)25 sqr(T x)
26 {
27     return x * x;
28 }
29 
main()30 int main()
31 {
32     {
33         typedef std::negative_binomial_distribution<> D;
34         typedef std::minstd_rand G;
35         G g;
36         D d(5, .25);
37         const int N = 1000000;
38         std::vector<D::result_type> u;
39         for (int i = 0; i < N; ++i)
40         {
41             D::result_type v = d(g);
42             assert(d.min() <= v && v <= d.max());
43             u.push_back(v);
44         }
45         double mean = std::accumulate(u.begin(), u.end(),
46                                               double(0)) / u.size();
47         double var = 0;
48         double skew = 0;
49         double kurtosis = 0;
50         for (int i = 0; i < u.size(); ++i)
51         {
52             double d = (u[i] - mean);
53             double d2 = sqr(d);
54             var += d2;
55             skew += d * d2;
56             kurtosis += d2 * d2;
57         }
58         var /= u.size();
59         double dev = std::sqrt(var);
60         skew /= u.size() * dev * var;
61         kurtosis /= u.size() * var * var;
62         kurtosis -= 3;
63         double x_mean = d.k() * (1 - d.p()) / d.p();
64         double x_var = x_mean / d.p();
65         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
66         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
67         assert(std::abs((mean - x_mean) / x_mean) < 0.01);
68         assert(std::abs((var - x_var) / x_var) < 0.01);
69         assert(std::abs((skew - x_skew) / x_skew) < 0.01);
70         assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.02);
71     }
72     {
73         typedef std::negative_binomial_distribution<> D;
74         typedef std::mt19937 G;
75         G g;
76         D d(30, .03125);
77         const int N = 1000000;
78         std::vector<D::result_type> u;
79         for (int i = 0; i < N; ++i)
80         {
81             D::result_type v = d(g);
82             assert(d.min() <= v && v <= d.max());
83             u.push_back(v);
84         }
85         double mean = std::accumulate(u.begin(), u.end(),
86                                               double(0)) / u.size();
87         double var = 0;
88         double skew = 0;
89         double kurtosis = 0;
90         for (int i = 0; i < u.size(); ++i)
91         {
92             double d = (u[i] - mean);
93             double d2 = sqr(d);
94             var += d2;
95             skew += d * d2;
96             kurtosis += d2 * d2;
97         }
98         var /= u.size();
99         double dev = std::sqrt(var);
100         skew /= u.size() * dev * var;
101         kurtosis /= u.size() * var * var;
102         kurtosis -= 3;
103         double x_mean = d.k() * (1 - d.p()) / d.p();
104         double x_var = x_mean / d.p();
105         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
106         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
107         assert(std::abs((mean - x_mean) / x_mean) < 0.01);
108         assert(std::abs((var - x_var) / x_var) < 0.01);
109         assert(std::abs((skew - x_skew) / x_skew) < 0.01);
110         assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.01);
111     }
112     {
113         typedef std::negative_binomial_distribution<> D;
114         typedef std::mt19937 G;
115         G g;
116         D d(40, .25);
117         const int N = 1000000;
118         std::vector<D::result_type> u;
119         for (int i = 0; i < N; ++i)
120         {
121             D::result_type v = d(g);
122             assert(d.min() <= v && v <= d.max());
123             u.push_back(v);
124         }
125         double mean = std::accumulate(u.begin(), u.end(),
126                                               double(0)) / u.size();
127         double var = 0;
128         double skew = 0;
129         double kurtosis = 0;
130         for (int i = 0; i < u.size(); ++i)
131         {
132             double d = (u[i] - mean);
133             double d2 = sqr(d);
134             var += d2;
135             skew += d * d2;
136             kurtosis += d2 * d2;
137         }
138         var /= u.size();
139         double dev = std::sqrt(var);
140         skew /= u.size() * dev * var;
141         kurtosis /= u.size() * var * var;
142         kurtosis -= 3;
143         double x_mean = d.k() * (1 - d.p()) / d.p();
144         double x_var = x_mean / d.p();
145         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
146         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
147         assert(std::abs((mean - x_mean) / x_mean) < 0.01);
148         assert(std::abs((var - x_var) / x_var) < 0.01);
149         assert(std::abs((skew - x_skew) / x_skew) < 0.01);
150         assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.03);
151     }
152     {
153         typedef std::negative_binomial_distribution<> D;
154         typedef std::mt19937 G;
155         G g;
156         D d(40, 1);
157         const int N = 1000;
158         std::vector<D::result_type> u;
159         for (int i = 0; i < N; ++i)
160         {
161             D::result_type v = d(g);
162             assert(d.min() <= v && v <= d.max());
163             u.push_back(v);
164         }
165         double mean = std::accumulate(u.begin(), u.end(),
166                                               double(0)) / u.size();
167         double var = 0;
168         double skew = 0;
169         double kurtosis = 0;
170         for (int i = 0; i < u.size(); ++i)
171         {
172             double d = (u[i] - mean);
173             double d2 = sqr(d);
174             var += d2;
175             skew += d * d2;
176             kurtosis += d2 * d2;
177         }
178         var /= u.size();
179         double dev = std::sqrt(var);
180         skew /= u.size() * dev * var;
181         kurtosis /= u.size() * var * var;
182         kurtosis -= 3;
183         double x_mean = d.k() * (1 - d.p()) / d.p();
184         double x_var = x_mean / d.p();
185         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
186         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
187         assert(mean == x_mean);
188         assert(var == x_var);
189     }
190     {
191         typedef std::negative_binomial_distribution<> D;
192         typedef std::mt19937 G;
193         G g;
194         D d(400, 0.5);
195         const int N = 1000000;
196         std::vector<D::result_type> u;
197         for (int i = 0; i < N; ++i)
198         {
199             D::result_type v = d(g);
200             assert(d.min() <= v && v <= d.max());
201             u.push_back(v);
202         }
203         double mean = std::accumulate(u.begin(), u.end(),
204                                               double(0)) / u.size();
205         double var = 0;
206         double skew = 0;
207         double kurtosis = 0;
208         for (int i = 0; i < u.size(); ++i)
209         {
210             double d = (u[i] - mean);
211             double d2 = sqr(d);
212             var += d2;
213             skew += d * d2;
214             kurtosis += d2 * d2;
215         }
216         var /= u.size();
217         double dev = std::sqrt(var);
218         skew /= u.size() * dev * var;
219         kurtosis /= u.size() * var * var;
220         kurtosis -= 3;
221         double x_mean = d.k() * (1 - d.p()) / d.p();
222         double x_var = x_mean / d.p();
223         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
224         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
225         assert(std::abs((mean - x_mean) / x_mean) < 0.01);
226         assert(std::abs((var - x_var) / x_var) < 0.01);
227         assert(std::abs((skew - x_skew) / x_skew) < 0.04);
228         assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.05);
229     }
230     {
231         typedef std::negative_binomial_distribution<> D;
232         typedef std::mt19937 G;
233         G g;
234         D d(1, 0.05);
235         const int N = 1000000;
236         std::vector<D::result_type> u;
237         for (int i = 0; i < N; ++i)
238         {
239             D::result_type v = d(g);
240             assert(d.min() <= v && v <= d.max());
241             u.push_back(v);
242         }
243         double mean = std::accumulate(u.begin(), u.end(),
244                                               double(0)) / u.size();
245         double var = 0;
246         double skew = 0;
247         double kurtosis = 0;
248         for (int i = 0; i < u.size(); ++i)
249         {
250             double d = (u[i] - mean);
251             double d2 = sqr(d);
252             var += d2;
253             skew += d * d2;
254             kurtosis += d2 * d2;
255         }
256         var /= u.size();
257         double dev = std::sqrt(var);
258         skew /= u.size() * dev * var;
259         kurtosis /= u.size() * var * var;
260         kurtosis -= 3;
261         double x_mean = d.k() * (1 - d.p()) / d.p();
262         double x_var = x_mean / d.p();
263         double x_skew = (2 - d.p()) / std::sqrt(d.k() * (1 - d.p()));
264         double x_kurtosis = 6. / d.k() + sqr(d.p()) / (d.k() * (1 - d.p()));
265         assert(std::abs((mean - x_mean) / x_mean) < 0.01);
266         assert(std::abs((var - x_var) / x_var) < 0.01);
267         assert(std::abs((skew - x_skew) / x_skew) < 0.01);
268         assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.03);
269     }
270 }
271