• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 ///////////////////////////////////////////////////////////////////////////////
2 // p_square_cumulative_distribution.hpp
3 //
4 //  Copyright 2005 Daniel Egloff, Olivier Gygi. Distributed under the Boost
5 //  Software License, Version 1.0. (See accompanying file
6 //  LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
7 
8 #ifndef BOOST_ACCUMULATORS_STATISTICS_P_SQUARE_CUMUL_DIST_HPP_DE_01_01_2006
9 #define BOOST_ACCUMULATORS_STATISTICS_P_SQUARE_CUMUL_DIST_HPP_DE_01_01_2006
10 
11 #include <vector>
12 #include <functional>
13 #include <boost/parameter/keyword.hpp>
14 #include <boost/range.hpp>
15 #include <boost/mpl/placeholders.hpp>
16 #include <boost/accumulators/accumulators_fwd.hpp>
17 #include <boost/accumulators/framework/accumulator_base.hpp>
18 #include <boost/accumulators/framework/extractor.hpp>
19 #include <boost/accumulators/numeric/functional.hpp>
20 #include <boost/accumulators/framework/parameters/sample.hpp>
21 #include <boost/accumulators/statistics_fwd.hpp>
22 #include <boost/accumulators/statistics/count.hpp>
23 #include <boost/serialization/vector.hpp>
24 #include <boost/serialization/utility.hpp>
25 
26 namespace boost { namespace accumulators
27 {
28 ///////////////////////////////////////////////////////////////////////////////
29 // num_cells named parameter
30 //
31 BOOST_PARAMETER_NESTED_KEYWORD(tag, p_square_cumulative_distribution_num_cells, num_cells)
32 
33 BOOST_ACCUMULATORS_IGNORE_GLOBAL(p_square_cumulative_distribution_num_cells)
34 
35 namespace impl
36 {
37     ///////////////////////////////////////////////////////////////////////////////
38     // p_square_cumulative_distribution_impl
39     //  cumulative_distribution calculation (as histogram)
40     /**
41         @brief Histogram calculation of the cumulative distribution with the \f$P^2\f$ algorithm
42 
43         A histogram of the sample cumulative distribution is computed dynamically without storing samples
44         based on the \f$ P^2 \f$ algorithm. The returned histogram has a specifiable amount (num_cells)
45         equiprobable (and not equal-sized) cells.
46 
47         For further details, see
48 
49         R. Jain and I. Chlamtac, The P^2 algorithm for dynamic calculation of quantiles and
50         histograms without storing observations, Communications of the ACM,
51         Volume 28 (October), Number 10, 1985, p. 1076-1085.
52 
53         @param p_square_cumulative_distribution_num_cells.
54     */
55     template<typename Sample>
56     struct p_square_cumulative_distribution_impl
57       : accumulator_base
58     {
59         typedef typename numeric::functional::fdiv<Sample, std::size_t>::result_type float_type;
60         typedef std::vector<float_type> array_type;
61         typedef std::vector<std::pair<float_type, float_type> > histogram_type;
62         // for boost::result_of
63         typedef iterator_range<typename histogram_type::iterator> result_type;
64 
65         template<typename Args>
p_square_cumulative_distribution_implboost::accumulators::impl::p_square_cumulative_distribution_impl66         p_square_cumulative_distribution_impl(Args const &args)
67           : num_cells(args[p_square_cumulative_distribution_num_cells])
68           , heights(num_cells + 1)
69           , actual_positions(num_cells + 1)
70           , desired_positions(num_cells + 1)
71           , positions_increments(num_cells + 1)
72           , histogram(num_cells + 1)
73           , is_dirty(true)
74         {
75             std::size_t b = this->num_cells;
76 
77             for (std::size_t i = 0; i < b + 1; ++i)
78             {
79                 this->actual_positions[i] = i + 1.;
80                 this->desired_positions[i] = i + 1.;
81                 this->positions_increments[i] = numeric::fdiv(i, b);
82             }
83         }
84 
85         template<typename Args>
operator ()boost::accumulators::impl::p_square_cumulative_distribution_impl86         void operator ()(Args const &args)
87         {
88             this->is_dirty = true;
89 
90             std::size_t cnt = count(args);
91             std::size_t sample_cell = 1; // k
92             std::size_t b = this->num_cells;
93 
94             // accumulate num_cells + 1 first samples
95             if (cnt <= b + 1)
96             {
97                 this->heights[cnt - 1] = args[sample];
98 
99                 // complete the initialization of heights by sorting
100                 if (cnt == b + 1)
101                 {
102                     std::sort(this->heights.begin(), this->heights.end());
103                 }
104             }
105             else
106             {
107                 // find cell k such that heights[k-1] <= args[sample] < heights[k] and adjust extreme values
108                 if (args[sample] < this->heights[0])
109                 {
110                     this->heights[0] = args[sample];
111                     sample_cell = 1;
112                 }
113                 else if (this->heights[b] <= args[sample])
114                 {
115                     this->heights[b] = args[sample];
116                     sample_cell = b;
117                 }
118                 else
119                 {
120                     typename array_type::iterator it;
121                     it = std::upper_bound(
122                         this->heights.begin()
123                       , this->heights.end()
124                       , args[sample]
125                     );
126 
127                     sample_cell = std::distance(this->heights.begin(), it);
128                 }
129 
130                 // increment positions of markers above sample_cell
131                 for (std::size_t i = sample_cell; i < b + 1; ++i)
132                 {
133                     ++this->actual_positions[i];
134                 }
135 
136                 // update desired position of markers 2 to num_cells + 1
137                 // (desired position of first marker is always 1)
138                 for (std::size_t i = 1; i < b + 1; ++i)
139                 {
140                     this->desired_positions[i] += this->positions_increments[i];
141                 }
142 
143                 // adjust heights of markers 2 to num_cells if necessary
144                 for (std::size_t i = 1; i < b; ++i)
145                 {
146                     // offset to desire position
147                     float_type d = this->desired_positions[i] - this->actual_positions[i];
148 
149                     // offset to next position
150                     float_type dp = this->actual_positions[i + 1] - this->actual_positions[i];
151 
152                     // offset to previous position
153                     float_type dm = this->actual_positions[i - 1] - this->actual_positions[i];
154 
155                     // height ds
156                     float_type hp = (this->heights[i + 1] - this->heights[i]) / dp;
157                     float_type hm = (this->heights[i - 1] - this->heights[i]) / dm;
158 
159                     if ( ( d >= 1. && dp > 1. ) || ( d <= -1. && dm < -1. ) )
160                     {
161                         short sign_d = static_cast<short>(d / std::abs(d));
162 
163                         // try adjusting heights[i] using p-squared formula
164                         float_type h = this->heights[i] + sign_d / (dp - dm) * ( (sign_d - dm) * hp + (dp - sign_d) * hm );
165 
166                         if ( this->heights[i - 1] < h && h < this->heights[i + 1] )
167                         {
168                             this->heights[i] = h;
169                         }
170                         else
171                         {
172                             // use linear formula
173                             if (d>0)
174                             {
175                                 this->heights[i] += hp;
176                             }
177                             if (d<0)
178                             {
179                                 this->heights[i] -= hm;
180                             }
181                         }
182                         this->actual_positions[i] += sign_d;
183                     }
184                 }
185             }
186         }
187 
188         template<typename Args>
resultboost::accumulators::impl::p_square_cumulative_distribution_impl189         result_type result(Args const &args) const
190         {
191             if (this->is_dirty)
192             {
193                 this->is_dirty = false;
194 
195                 // creates a vector of std::pair where each pair i holds
196                 // the values heights[i] (x-axis of histogram) and
197                 // actual_positions[i] / cnt (y-axis of histogram)
198 
199                 std::size_t cnt = count(args);
200 
201                 for (std::size_t i = 0; i < this->histogram.size(); ++i)
202                 {
203                     this->histogram[i] = std::make_pair(this->heights[i], numeric::fdiv(this->actual_positions[i], cnt));
204                 }
205             }
206             //return histogram;
207             return make_iterator_range(this->histogram);
208         }
209 
210         // make this accumulator serializeable
211         // TODO split to save/load and check on parameters provided in ctor
212         template<class Archive>
serializeboost::accumulators::impl::p_square_cumulative_distribution_impl213         void serialize(Archive & ar, const unsigned int file_version)
214         {
215             ar & num_cells;
216             ar & heights;
217             ar & actual_positions;
218             ar & desired_positions;
219             ar & positions_increments;
220             ar & histogram;
221             ar & is_dirty;
222         }
223 
224     private:
225         std::size_t num_cells;            // number of cells b
226         array_type  heights;              // q_i
227         array_type  actual_positions;     // n_i
228         array_type  desired_positions;    // n'_i
229         array_type  positions_increments; // dn'_i
230         mutable histogram_type histogram; // histogram
231         mutable bool is_dirty;
232     };
233 
234 } // namespace detail
235 
236 ///////////////////////////////////////////////////////////////////////////////
237 // tag::p_square_cumulative_distribution
238 //
239 namespace tag
240 {
241     struct p_square_cumulative_distribution
242       : depends_on<count>
243       , p_square_cumulative_distribution_num_cells
244     {
245         /// INTERNAL ONLY
246         ///
247         typedef accumulators::impl::p_square_cumulative_distribution_impl<mpl::_1> impl;
248     };
249 }
250 
251 ///////////////////////////////////////////////////////////////////////////////
252 // extract::p_square_cumulative_distribution
253 //
254 namespace extract
255 {
256     extractor<tag::p_square_cumulative_distribution> const p_square_cumulative_distribution = {};
257 
258     BOOST_ACCUMULATORS_IGNORE_GLOBAL(p_square_cumulative_distribution)
259 }
260 
261 using extract::p_square_cumulative_distribution;
262 
263 // So that p_square_cumulative_distribution can be automatically substituted with
264 // weighted_p_square_cumulative_distribution when the weight parameter is non-void
265 template<>
266 struct as_weighted_feature<tag::p_square_cumulative_distribution>
267 {
268     typedef tag::weighted_p_square_cumulative_distribution type;
269 };
270 
271 template<>
272 struct feature_of<tag::weighted_p_square_cumulative_distribution>
273   : feature_of<tag::p_square_cumulative_distribution>
274 {
275 };
276 
277 }} // namespace boost::accumulators
278 
279 #endif
280