xref: /aosp_15_r20/external/angle/third_party/abseil-cpp/absl/random/discrete_distribution.h (revision 8975f5c5ed3d1c378011245431ada316dfb6f244)
1 // Copyright 2017 The Abseil Authors.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //      https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #ifndef ABSL_RANDOM_DISCRETE_DISTRIBUTION_H_
16 #define ABSL_RANDOM_DISCRETE_DISTRIBUTION_H_
17 
18 #include <cassert>
19 #include <cstddef>
20 #include <initializer_list>
21 #include <istream>
22 #include <limits>
23 #include <ostream>
24 #include <type_traits>
25 #include <utility>
26 #include <vector>
27 
28 #include "absl/base/config.h"
29 #include "absl/random/bernoulli_distribution.h"
30 #include "absl/random/internal/iostream_state_saver.h"
31 #include "absl/random/uniform_int_distribution.h"
32 
33 namespace absl {
34 ABSL_NAMESPACE_BEGIN
35 
36 // absl::discrete_distribution
37 //
38 // A discrete distribution produces random integers i, where 0 <= i < n
39 // distributed according to the discrete probability function:
40 //
41 //     P(i|p0,...,pn−1)=pi
42 //
43 // This class is an implementation of discrete_distribution (see
44 // [rand.dist.samp.discrete]).
45 //
46 // The algorithm used is Walker's Aliasing algorithm, described in Knuth, Vol 2.
47 // absl::discrete_distribution takes O(N) time to precompute the probabilities
48 // (where N is the number of possible outcomes in the distribution) at
49 // construction, and then takes O(1) time for each variate generation.  Many
50 // other implementations also take O(N) time to construct an ordered sequence of
51 // partial sums, plus O(log N) time per variate to binary search.
52 //
53 template <typename IntType = int>
54 class discrete_distribution {
55  public:
56   using result_type = IntType;
57 
58   class param_type {
59    public:
60     using distribution_type = discrete_distribution;
61 
param_type()62     param_type() { init(); }
63 
64     template <typename InputIterator>
param_type(InputIterator begin,InputIterator end)65     explicit param_type(InputIterator begin, InputIterator end)
66         : p_(begin, end) {
67       init();
68     }
69 
param_type(std::initializer_list<double> weights)70     explicit param_type(std::initializer_list<double> weights) : p_(weights) {
71       init();
72     }
73 
74     template <class UnaryOperation>
param_type(size_t nw,double xmin,double xmax,UnaryOperation fw)75     explicit param_type(size_t nw, double xmin, double xmax,
76                         UnaryOperation fw) {
77       if (nw > 0) {
78         p_.reserve(nw);
79         double delta = (xmax - xmin) / static_cast<double>(nw);
80         assert(delta > 0);
81         double t = delta * 0.5;
82         for (size_t i = 0; i < nw; ++i) {
83           p_.push_back(fw(xmin + i * delta + t));
84         }
85       }
86       init();
87     }
88 
probabilities()89     const std::vector<double>& probabilities() const { return p_; }
n()90     size_t n() const { return p_.size() - 1; }
91 
92     friend bool operator==(const param_type& a, const param_type& b) {
93       return a.probabilities() == b.probabilities();
94     }
95 
96     friend bool operator!=(const param_type& a, const param_type& b) {
97       return !(a == b);
98     }
99 
100    private:
101     friend class discrete_distribution;
102 
103     void init();
104 
105     std::vector<double> p_;                     // normalized probabilities
106     std::vector<std::pair<double, size_t>> q_;  // (acceptance, alternate) pairs
107 
108     static_assert(std::is_integral<result_type>::value,
109                   "Class-template absl::discrete_distribution<> must be "
110                   "parameterized using an integral type.");
111   };
112 
discrete_distribution()113   discrete_distribution() : param_() {}
114 
discrete_distribution(const param_type & p)115   explicit discrete_distribution(const param_type& p) : param_(p) {}
116 
117   template <typename InputIterator>
discrete_distribution(InputIterator begin,InputIterator end)118   explicit discrete_distribution(InputIterator begin, InputIterator end)
119       : param_(begin, end) {}
120 
discrete_distribution(std::initializer_list<double> weights)121   explicit discrete_distribution(std::initializer_list<double> weights)
122       : param_(weights) {}
123 
124   template <class UnaryOperation>
discrete_distribution(size_t nw,double xmin,double xmax,UnaryOperation fw)125   explicit discrete_distribution(size_t nw, double xmin, double xmax,
126                                  UnaryOperation fw)
127       : param_(nw, xmin, xmax, std::move(fw)) {}
128 
reset()129   void reset() {}
130 
131   // generating functions
132   template <typename URBG>
operator()133   result_type operator()(URBG& g) {  // NOLINT(runtime/references)
134     return (*this)(g, param_);
135   }
136 
137   template <typename URBG>
138   result_type operator()(URBG& g,  // NOLINT(runtime/references)
139                          const param_type& p);
140 
param()141   const param_type& param() const { return param_; }
param(const param_type & p)142   void param(const param_type& p) { param_ = p; }
143 
result_type(min)144   result_type(min)() const { return 0; }
result_type(max)145   result_type(max)() const {
146     return static_cast<result_type>(param_.n());
147   }  // inclusive
148 
149   // NOTE [rand.dist.sample.discrete] returns a std::vector<double> not a
150   // const std::vector<double>&.
probabilities()151   const std::vector<double>& probabilities() const {
152     return param_.probabilities();
153   }
154 
155   friend bool operator==(const discrete_distribution& a,
156                          const discrete_distribution& b) {
157     return a.param_ == b.param_;
158   }
159   friend bool operator!=(const discrete_distribution& a,
160                          const discrete_distribution& b) {
161     return a.param_ != b.param_;
162   }
163 
164  private:
165   param_type param_;
166 };
167 
168 // --------------------------------------------------------------------------
169 // Implementation details only below
170 // --------------------------------------------------------------------------
171 
172 namespace random_internal {
173 
174 // Using the vector `*probabilities`, whose values are the weights or
175 // probabilities of an element being selected, constructs the proportional
176 // probabilities used by the discrete distribution.  `*probabilities` will be
177 // scaled, if necessary, so that its entries sum to a value sufficiently close
178 // to 1.0.
179 std::vector<std::pair<double, size_t>> InitDiscreteDistribution(
180     std::vector<double>* probabilities);
181 
182 }  // namespace random_internal
183 
184 template <typename IntType>
init()185 void discrete_distribution<IntType>::param_type::init() {
186   if (p_.empty()) {
187     p_.push_back(1.0);
188     q_.emplace_back(1.0, 0);
189   } else {
190     assert(n() <= (std::numeric_limits<IntType>::max)());
191     q_ = random_internal::InitDiscreteDistribution(&p_);
192   }
193 }
194 
195 template <typename IntType>
196 template <typename URBG>
197 typename discrete_distribution<IntType>::result_type
operator()198 discrete_distribution<IntType>::operator()(
199     URBG& g,  // NOLINT(runtime/references)
200     const param_type& p) {
201   const auto idx = absl::uniform_int_distribution<result_type>(0, p.n())(g);
202   const auto& q = p.q_[idx];
203   const bool selected = absl::bernoulli_distribution(q.first)(g);
204   return selected ? idx : static_cast<result_type>(q.second);
205 }
206 
207 template <typename CharT, typename Traits, typename IntType>
208 std::basic_ostream<CharT, Traits>& operator<<(
209     std::basic_ostream<CharT, Traits>& os,  // NOLINT(runtime/references)
210     const discrete_distribution<IntType>& x) {
211   auto saver = random_internal::make_ostream_state_saver(os);
212   const auto& probabilities = x.param().probabilities();
213   os << probabilities.size();
214 
215   os.precision(random_internal::stream_precision_helper<double>::kPrecision);
216   for (const auto& p : probabilities) {
217     os << os.fill() << p;
218   }
219   return os;
220 }
221 
222 template <typename CharT, typename Traits, typename IntType>
223 std::basic_istream<CharT, Traits>& operator>>(
224     std::basic_istream<CharT, Traits>& is,  // NOLINT(runtime/references)
225     discrete_distribution<IntType>& x) {    // NOLINT(runtime/references)
226   using param_type = typename discrete_distribution<IntType>::param_type;
227   auto saver = random_internal::make_istream_state_saver(is);
228 
229   size_t n;
230   std::vector<double> p;
231 
232   is >> n;
233   if (is.fail()) return is;
234   if (n > 0) {
235     p.reserve(n);
236     for (IntType i = 0; i < n && !is.fail(); ++i) {
237       auto tmp = random_internal::read_floating_point<double>(is);
238       if (is.fail()) return is;
239       p.push_back(tmp);
240     }
241   }
242   x.param(param_type(p.begin(), p.end()));
243   return is;
244 }
245 
246 ABSL_NAMESPACE_END
247 }  // namespace absl
248 
249 #endif  // ABSL_RANDOM_DISCRETE_DISTRIBUTION_H_
250