Thrill  0.1
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
bernoulli_sample.hpp
Go to the documentation of this file.
1 /*******************************************************************************
2  * thrill/api/bernoulli_sample.hpp
3  *
4  * Part of Project Thrill - http://project-thrill.org
5  *
6  * Copyright (C) 2016 Lorenz Hübschle-Schneider <[email protected]>
7  *
8  * All rights reserved. Published under the BSD-2 license in the LICENSE file.
9  ******************************************************************************/
10 
11 #pragma once
12 #ifndef THRILL_API_BERNOULLI_SAMPLE_HEADER
13 #define THRILL_API_BERNOULLI_SAMPLE_HEADER
14 
15 #include <thrill/api/dia.hpp>
17 
18 #include <random>
19 
20 namespace thrill {
21 namespace api {
22 
23 /*!
24  * \ingroup api_layer
25  */
26 template <typename ValueType>
28 {
29  static const bool debug = false;
30 
31  using SkipDistValueType = int;
32 
33 public:
34  explicit BernoulliSampleNode(double p)
35  : p_(p), use_skip_(p < 0.1) { // use skip values if p < 0.1
36  assert(p >= 0.0 && p <= 1.0);
37 
38  if (use_skip_) {
39  skip_dist_ = std::geometric_distribution<SkipDistValueType>(p);
41 
42  LOG << "Skip value initialised with " << skip_remaining_;
43  }
44  else {
45  simple_dist_ = std::bernoulli_distribution(p);
46  }
47  }
48 
49  template <typename Emitter>
50  inline void operator () (const ValueType& item, Emitter&& emit) {
51  if (use_skip_) {
52  // use geometric distribution and skip values
53  if (skip_remaining_ == 0) {
54  // sample element
55  LOG << "sampled item " << item;
56  emit(item);
58  }
59  else {
61  }
62  }
63  else {
64  // use bernoulli distribution
65  if (simple_dist_(rng_)) {
66  LOG << "sampled item " << item;
67  emit(item);
68  }
69  }
70  }
71 
72  bool use_skip() const {
73  return use_skip_;
74  }
75 
76 private:
77  // Sampling rate
78  const double p_;
79  // Whether to generate skip values with a geometric distribution or to use
80  // the naive method
81  const bool use_skip_;
82  // Random generator
83  std::default_random_engine rng_ { std::random_device { } () };
84  std::bernoulli_distribution simple_dist_;
85  std::geometric_distribution<SkipDistValueType> skip_dist_;
87 };
88 
89 template <typename ValueType, typename Stack>
90 auto DIA<ValueType, Stack>::BernoulliSample(const double p) const {
91  assert(IsValid());
92 
93  size_t new_id = context().next_dia_id();
94 
95  node_->context().logger_
96  << "id" << new_id
97  << "label" << "BernoulliSample"
98  << "class" << "DIA"
99  << "event" << "create"
100  << "type" << "LOp"
101  << "parents" << (common::Array<size_t>{ id_ });
102 
103  auto new_stack = stack_.push(BernoulliSampleNode<ValueType>(p));
105  node_, new_stack, new_id, "BernoulliSample");
106 }
107 
108 } // namespace api
109 } // namespace thrill
110 
111 #endif // !THRILL_API_BERNOULLI_SAMPLE_HEADER
112 
113 /******************************************************************************/
DIA is the interface between the user and the Thrill framework.
Definition: dia.hpp:141
Type[] Array
A template to make writing temporary arrays easy: Array<int>{ 1, 2, 3 }.
Definition: json_logger.hpp:64
std::default_random_engine rng_
std::geometric_distribution< SkipDistValueType > skip_dist_
auto BernoulliSample(double p) const
Each item of a DIA is copied into the output DIA with success probability p (an independent Bernoulli...
#define LOG
Default logging method: output if the local debug variable is true.
Definition: logger.hpp:24
std::bernoulli_distribution simple_dist_
void operator()(const ValueType &item, Emitter &&emit)