1 //===----------------------------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #ifndef _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
10 #define _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
11 
12 #include <__config>
13 #include <__random/clamp_to_integral.h>
14 #include <__random/exponential_distribution.h>
15 #include <__random/is_valid.h>
16 #include <__random/normal_distribution.h>
17 #include <__random/uniform_real_distribution.h>
18 #include <cmath>
19 #include <iosfwd>
20 #include <limits>
21 
22 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
23 #  pragma GCC system_header
24 #endif
25 
26 _LIBCPP_PUSH_MACROS
27 #include <__undef_macros>
28 
29 _LIBCPP_BEGIN_NAMESPACE_STD
30 
31 template<class _IntType = int>
32 class _LIBCPP_TEMPLATE_VIS poisson_distribution
33 {
34     static_assert(__libcpp_random_is_valid_inttype<_IntType>::value, "IntType must be a supported integer type");
35 public:
36     // types
37     typedef _IntType result_type;
38 
39     class _LIBCPP_TEMPLATE_VIS param_type
40     {
41         double __mean_;
42         double __s_;
43         double __d_;
44         double __l_;
45         double __omega_;
46         double __c0_;
47         double __c1_;
48         double __c2_;
49         double __c3_;
50         double __c_;
51 
52     public:
53         typedef poisson_distribution distribution_type;
54 
55         _LIBCPP_HIDE_FROM_ABI explicit param_type(double __mean = 1.0);
56 
57         _LIBCPP_INLINE_VISIBILITY
58         double mean() const {return __mean_;}
59 
60         friend _LIBCPP_INLINE_VISIBILITY
61             bool operator==(const param_type& __x, const param_type& __y)
62             {return __x.__mean_ == __y.__mean_;}
63         friend _LIBCPP_INLINE_VISIBILITY
64             bool operator!=(const param_type& __x, const param_type& __y)
65             {return !(__x == __y);}
66 
67         friend class poisson_distribution;
68     };
69 
70 private:
71     param_type __p_;
72 
73 public:
74     // constructors and reset functions
75 #ifndef _LIBCPP_CXX03_LANG
76     _LIBCPP_INLINE_VISIBILITY
77     poisson_distribution() : poisson_distribution(1.0) {}
78     _LIBCPP_INLINE_VISIBILITY
79     explicit poisson_distribution(double __mean)
80         : __p_(__mean) {}
81 #else
82     _LIBCPP_INLINE_VISIBILITY
83     explicit poisson_distribution(double __mean = 1.0)
84         : __p_(__mean) {}
85 #endif
86     _LIBCPP_INLINE_VISIBILITY
87     explicit poisson_distribution(const param_type& __p) : __p_(__p) {}
88     _LIBCPP_INLINE_VISIBILITY
89     void reset() {}
90 
91     // generating functions
92     template<class _URNG>
93         _LIBCPP_INLINE_VISIBILITY
94         result_type operator()(_URNG& __g)
95         {return (*this)(__g, __p_);}
96     template<class _URNG>
97     _LIBCPP_HIDE_FROM_ABI result_type operator()(_URNG& __g, const param_type& __p);
98 
99     // property functions
100     _LIBCPP_INLINE_VISIBILITY
101     double mean() const {return __p_.mean();}
102 
103     _LIBCPP_INLINE_VISIBILITY
104     param_type param() const {return __p_;}
105     _LIBCPP_INLINE_VISIBILITY
106     void param(const param_type& __p) {__p_ = __p;}
107 
108     _LIBCPP_INLINE_VISIBILITY
109     result_type min() const {return 0;}
110     _LIBCPP_INLINE_VISIBILITY
111     result_type max() const {return numeric_limits<result_type>::max();}
112 
113     friend _LIBCPP_INLINE_VISIBILITY
114         bool operator==(const poisson_distribution& __x,
115                         const poisson_distribution& __y)
116         {return __x.__p_ == __y.__p_;}
117     friend _LIBCPP_INLINE_VISIBILITY
118         bool operator!=(const poisson_distribution& __x,
119                         const poisson_distribution& __y)
120         {return !(__x == __y);}
121 };
122 
123 template<class _IntType>
124 poisson_distribution<_IntType>::param_type::param_type(double __mean)
125     // According to the standard `inf` is a valid input, but it causes the
126     // distribution to hang, so we replace it with the maximum representable
127     // mean.
128     : __mean_(isinf(__mean) ? numeric_limits<double>::max() : __mean)
129 {
130     if (__mean_ < 10)
131     {
132         __s_ = 0;
133         __d_ = 0;
134         __l_ = _VSTD::exp(-__mean_);
135         __omega_ = 0;
136         __c3_ = 0;
137         __c2_ = 0;
138         __c1_ = 0;
139         __c0_ = 0;
140         __c_ = 0;
141     }
142     else
143     {
144         __s_ = _VSTD::sqrt(__mean_);
145         __d_ = 6 * __mean_ * __mean_;
146         __l_ = _VSTD::trunc(__mean_ - 1.1484);
147         __omega_ = .3989423 / __s_;
148         double __b1 = .4166667E-1 / __mean_;
149         double __b2 = .3 * __b1 * __b1;
150         __c3_ = .1428571 * __b1 * __b2;
151         __c2_ = __b2 - 15. * __c3_;
152         __c1_ = __b1 - 6. * __b2 + 45. * __c3_;
153         __c0_ = 1. - __b1 + 3. * __b2 - 15. * __c3_;
154         __c_ = .1069 / __mean_;
155     }
156 }
157 
158 template <class _IntType>
159 template<class _URNG>
160 _IntType
161 poisson_distribution<_IntType>::operator()(_URNG& __urng, const param_type& __pr)
162 {
163     static_assert(__libcpp_random_is_valid_urng<_URNG>::value, "");
164     double __tx;
165     uniform_real_distribution<double> __urd;
166     if (__pr.__mean_ < 10)
167     {
168          __tx = 0;
169         for (double __p = __urd(__urng); __p > __pr.__l_; ++__tx)
170             __p *= __urd(__urng);
171     }
172     else
173     {
174         double __difmuk;
175         double __g = __pr.__mean_ + __pr.__s_ * normal_distribution<double>()(__urng);
176         double __u;
177         if (__g > 0)
178         {
179             __tx = _VSTD::trunc(__g);
180             if (__tx >= __pr.__l_)
181                 return _VSTD::__clamp_to_integral<result_type>(__tx);
182             __difmuk = __pr.__mean_ - __tx;
183             __u = __urd(__urng);
184             if (__pr.__d_ * __u >= __difmuk * __difmuk * __difmuk)
185                 return _VSTD::__clamp_to_integral<result_type>(__tx);
186         }
187         exponential_distribution<double> __edist;
188         for (bool __using_exp_dist = false; true; __using_exp_dist = true)
189         {
190             double __e;
191             if (__using_exp_dist || __g <= 0)
192             {
193                 double __t;
194                 do
195                 {
196                     __e = __edist(__urng);
197                     __u = __urd(__urng);
198                     __u += __u - 1;
199                     __t = 1.8 + (__u < 0 ? -__e : __e);
200                 } while (__t <= -.6744);
201                 __tx = _VSTD::trunc(__pr.__mean_ + __pr.__s_ * __t);
202                 __difmuk = __pr.__mean_ - __tx;
203                 __using_exp_dist = true;
204             }
205             double __px;
206             double __py;
207             if (__tx < 10 && __tx >= 0)
208             {
209                 const double __fac[] = {1, 1, 2, 6, 24, 120, 720, 5040,
210                                              40320, 362880};
211                 __px = -__pr.__mean_;
212                 __py = _VSTD::pow(__pr.__mean_, (double)__tx) / __fac[static_cast<int>(__tx)];
213             }
214             else
215             {
216                 double __del = .8333333E-1 / __tx;
217                 __del -= 4.8 * __del * __del * __del;
218                 double __v = __difmuk / __tx;
219                 if (_VSTD::abs(__v) > 0.25)
220                     __px = __tx * _VSTD::log(1 + __v) - __difmuk - __del;
221                 else
222                     __px = __tx * __v * __v * (((((((.1250060 * __v + -.1384794) *
223                            __v + .1421878) * __v + -.1661269) * __v + .2000118) *
224                            __v + -.2500068) * __v + .3333333) * __v + -.5) - __del;
225                 __py = .3989423 / _VSTD::sqrt(__tx);
226             }
227             double __r = (0.5 - __difmuk) / __pr.__s_;
228             double __r2 = __r * __r;
229             double __fx = -0.5 * __r2;
230             double __fy = __pr.__omega_ * (((__pr.__c3_ * __r2 + __pr.__c2_) *
231                                         __r2 + __pr.__c1_) * __r2 + __pr.__c0_);
232             if (__using_exp_dist)
233             {
234                 if (__pr.__c_ * _VSTD::abs(__u) <= __py * _VSTD::exp(__px + __e) -
235                                                    __fy * _VSTD::exp(__fx + __e))
236                     break;
237             }
238             else
239             {
240                 if (__fy - __u * __fy <= __py * _VSTD::exp(__px - __fx))
241                     break;
242             }
243         }
244     }
245     return _VSTD::__clamp_to_integral<result_type>(__tx);
246 }
247 
248 template <class _CharT, class _Traits, class _IntType>
249 _LIBCPP_HIDE_FROM_ABI basic_ostream<_CharT, _Traits>&
250 operator<<(basic_ostream<_CharT, _Traits>& __os,
251            const poisson_distribution<_IntType>& __x)
252 {
253     __save_flags<_CharT, _Traits> __lx(__os);
254     typedef basic_ostream<_CharT, _Traits> _OStream;
255     __os.flags(_OStream::dec | _OStream::left | _OStream::fixed |
256                _OStream::scientific);
257     return __os << __x.mean();
258 }
259 
260 template <class _CharT, class _Traits, class _IntType>
261 _LIBCPP_HIDE_FROM_ABI basic_istream<_CharT, _Traits>&
262 operator>>(basic_istream<_CharT, _Traits>& __is,
263            poisson_distribution<_IntType>& __x)
264 {
265     typedef poisson_distribution<_IntType> _Eng;
266     typedef typename _Eng::param_type param_type;
267     __save_flags<_CharT, _Traits> __lx(__is);
268     typedef basic_istream<_CharT, _Traits> _Istream;
269     __is.flags(_Istream::dec | _Istream::skipws);
270     double __mean;
271     __is >> __mean;
272     if (!__is.fail())
273         __x.param(param_type(__mean));
274     return __is;
275 }
276 
277 _LIBCPP_END_NAMESPACE_STD
278 
279 _LIBCPP_POP_MACROS
280 
281 #endif // _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
282