/usr/include/dlib/svm/ranking_tools.h is in libdlib-dev 18.18-2build1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 | // Copyright (C) 2012 Davis E. King (davis@dlib.net)
// License: Boost Software License See LICENSE.txt for the full license.
#ifndef DLIB_RANKING_ToOLS_Hh_
#define DLIB_RANKING_ToOLS_Hh_
#include "ranking_tools_abstract.h"
#include "../algs.h"
#include "../matrix.h"
#include <vector>
#include <utility>
#include <algorithm>
#include "sparse_vector.h"
#include "../statistics.h"
namespace dlib
{
// ----------------------------------------------------------------------------------------
template <
typename T
>
struct ranking_pair
{
ranking_pair() {}
ranking_pair(
const std::vector<T>& r,
const std::vector<T>& nr
) :
relevant(r), nonrelevant(nr)
{}
std::vector<T> relevant;
std::vector<T> nonrelevant;
};
template <
typename T
>
void serialize (
const ranking_pair<T>& item,
std::ostream& out
)
{
int version = 1;
serialize(version, out);
serialize(item.relevant, out);
serialize(item.nonrelevant, out);
}
template <
typename T
>
void deserialize (
ranking_pair<T>& item,
std::istream& in
)
{
int version = 0;
deserialize(version, in);
if (version != 1)
throw dlib::serialization_error("Wrong version found while deserializing dlib::ranking_pair");
deserialize(item.relevant, in);
deserialize(item.nonrelevant, in);
}
// ----------------------------------------------------------------------------------------
template <
typename T
>
typename disable_if<is_matrix<T>,bool>::type is_ranking_problem (
const std::vector<ranking_pair<T> >& samples
)
{
if (samples.size() == 0)
return false;
for (unsigned long i = 0; i < samples.size(); ++i)
{
if (samples[i].relevant.size() == 0)
return false;
if (samples[i].nonrelevant.size() == 0)
return false;
}
return true;
}
template <
typename T
>
typename enable_if<is_matrix<T>,bool>::type is_ranking_problem (
const std::vector<ranking_pair<T> >& samples
)
{
if (samples.size() == 0)
return false;
for (unsigned long i = 0; i < samples.size(); ++i)
{
if (samples[i].relevant.size() == 0)
return false;
if (samples[i].nonrelevant.size() == 0)
return false;
}
// If these are dense vectors then they must all have the same dimensionality.
const long dims = max_index_plus_one(samples[0].relevant);
for (unsigned long i = 0; i < samples.size(); ++i)
{
for (unsigned long j = 0; j < samples[i].relevant.size(); ++j)
{
if (is_vector(samples[i].relevant[j]) == false)
return false;
if (samples[i].relevant[j].size() != dims)
return false;
}
for (unsigned long j = 0; j < samples[i].nonrelevant.size(); ++j)
{
if (is_vector(samples[i].nonrelevant[j]) == false)
return false;
if (samples[i].nonrelevant[j].size() != dims)
return false;
}
}
return true;
}
// ----------------------------------------------------------------------------------------
template <
typename T
>
unsigned long max_index_plus_one (
const ranking_pair<T>& item
)
{
return std::max(max_index_plus_one(item.relevant), max_index_plus_one(item.nonrelevant));
}
template <
typename T
>
unsigned long max_index_plus_one (
const std::vector<ranking_pair<T> >& samples
)
{
unsigned long dims = 0;
for (unsigned long i = 0; i < samples.size(); ++i)
{
dims = std::max(dims, max_index_plus_one(samples[i]));
}
return dims;
}
// ----------------------------------------------------------------------------------------
template <typename T>
void count_ranking_inversions (
const std::vector<T>& x,
const std::vector<T>& y,
std::vector<unsigned long>& x_count,
std::vector<unsigned long>& y_count
)
{
x_count.assign(x.size(),0);
y_count.assign(y.size(),0);
if (x.size() == 0 || y.size() == 0)
return;
std::vector<std::pair<T,unsigned long> > xsort(x.size());
std::vector<std::pair<T,unsigned long> > ysort(y.size());
for (unsigned long i = 0; i < x.size(); ++i)
xsort[i] = std::make_pair(x[i], i);
for (unsigned long j = 0; j < y.size(); ++j)
ysort[j] = std::make_pair(y[j], j);
std::sort(xsort.begin(), xsort.end());
std::sort(ysort.begin(), ysort.end());
unsigned long i, j;
// Do the counting for the x values.
for (i = 0, j = 0; i < x_count.size(); ++i)
{
// Skip past y values that are in the correct order with respect to xsort[i].
while (j < ysort.size() && ysort[j].first < xsort[i].first)
++j;
x_count[xsort[i].second] = ysort.size() - j;
}
// Now do the counting for the y values.
for (i = 0, j = 0; j < y_count.size(); ++j)
{
// Skip past x values that are in the incorrect order with respect to ysort[j].
while (i < xsort.size() && !(ysort[j].first < xsort[i].first))
++i;
y_count[ysort[j].second] = i;
}
}
// ----------------------------------------------------------------------------------------
namespace impl
{
inline bool compare_first_reverse_second (
const std::pair<double,bool>& a,
const std::pair<double,bool>& b
)
{
if (a.first < b.first)
return true;
else if (a.first > b.first)
return false;
else if (a.second && !b.second)
return true;
else
return false;
}
}
template <
typename ranking_function,
typename T
>
matrix<double,1,2> test_ranking_function (
const ranking_function& funct,
const std::vector<ranking_pair<T> >& samples
)
{
// make sure requires clause is not broken
DLIB_ASSERT(is_ranking_problem(samples),
"\t double test_ranking_function()"
<< "\n\t invalid inputs were given to this function"
<< "\n\t samples.size(): " << samples.size()
<< "\n\t is_ranking_problem(samples): " << is_ranking_problem(samples)
);
unsigned long total_pairs = 0;
unsigned long total_wrong = 0;
std::vector<double> rel_scores;
std::vector<double> nonrel_scores;
std::vector<unsigned long> rel_counts;
std::vector<unsigned long> nonrel_counts;
running_stats<double> rs;
std::vector<std::pair<double,bool> > total_scores;
std::vector<bool> total_ranking;
for (unsigned long i = 0; i < samples.size(); ++i)
{
rel_scores.resize(samples[i].relevant.size());
nonrel_scores.resize(samples[i].nonrelevant.size());
total_scores.clear();
for (unsigned long k = 0; k < rel_scores.size(); ++k)
{
rel_scores[k] = funct(samples[i].relevant[k]);
total_scores.push_back(std::make_pair(rel_scores[k], true));
}
for (unsigned long k = 0; k < nonrel_scores.size(); ++k)
{
nonrel_scores[k] = funct(samples[i].nonrelevant[k]);
total_scores.push_back(std::make_pair(nonrel_scores[k], false));
}
// Now compute the average precision for this sample. We need to sort the
// results and the back them into total_ranking. Note that we sort them so
// that, if you get a block of ranking values that are all equal, the elements
// marked as true will come last. This prevents a ranking from outputting a
// constant value for everything and still getting a good MAP score.
std::sort(total_scores.rbegin(), total_scores.rend(), impl::compare_first_reverse_second);
total_ranking.clear();
for (unsigned long i = 0; i < total_scores.size(); ++i)
total_ranking.push_back(total_scores[i].second);
rs.add(average_precision(total_ranking));
count_ranking_inversions(rel_scores, nonrel_scores, rel_counts, nonrel_counts);
total_pairs += rel_scores.size()*nonrel_scores.size();
// Note that we don't need to look at nonrel_counts since it is redundant with
// the information in rel_counts in this case.
total_wrong += sum(mat(rel_counts));
}
const double rank_swaps = static_cast<double>(total_pairs - total_wrong) / total_pairs;
const double mean_average_precision = rs.mean();
matrix<double,1,2> res;
res = rank_swaps, mean_average_precision;
return res;
}
// ----------------------------------------------------------------------------------------
template <
typename ranking_function,
typename T
>
matrix<double,1,2> test_ranking_function (
const ranking_function& funct,
const ranking_pair<T>& sample
)
{
return test_ranking_function(funct, std::vector<ranking_pair<T> >(1,sample));
}
// ----------------------------------------------------------------------------------------
template <
typename trainer_type,
typename T
>
matrix<double,1,2> cross_validate_ranking_trainer (
const trainer_type& trainer,
const std::vector<ranking_pair<T> >& samples,
const long folds
)
{
// make sure requires clause is not broken
DLIB_ASSERT(is_ranking_problem(samples) &&
1 < folds && folds <= static_cast<long>(samples.size()),
"\t double cross_validate_ranking_trainer()"
<< "\n\t invalid inputs were given to this function"
<< "\n\t samples.size(): " << samples.size()
<< "\n\t folds: " << folds
<< "\n\t is_ranking_problem(samples): " << is_ranking_problem(samples)
);
const long num_in_test = samples.size()/folds;
const long num_in_train = samples.size() - num_in_test;
std::vector<ranking_pair<T> > samples_test, samples_train;
long next_test_idx = 0;
unsigned long total_pairs = 0;
unsigned long total_wrong = 0;
std::vector<double> rel_scores;
std::vector<double> nonrel_scores;
std::vector<unsigned long> rel_counts;
std::vector<unsigned long> nonrel_counts;
running_stats<double> rs;
std::vector<std::pair<double,bool> > total_scores;
std::vector<bool> total_ranking;
for (long i = 0; i < folds; ++i)
{
samples_test.clear();
samples_train.clear();
// load up the test samples
for (long cnt = 0; cnt < num_in_test; ++cnt)
{
samples_test.push_back(samples[next_test_idx]);
next_test_idx = (next_test_idx + 1)%samples.size();
}
// load up the training samples
long next = next_test_idx;
for (long cnt = 0; cnt < num_in_train; ++cnt)
{
samples_train.push_back(samples[next]);
next = (next + 1)%samples.size();
}
const typename trainer_type::trained_function_type& df = trainer.train(samples_train);
// check how good df is on the test data
for (unsigned long i = 0; i < samples_test.size(); ++i)
{
rel_scores.resize(samples_test[i].relevant.size());
nonrel_scores.resize(samples_test[i].nonrelevant.size());
total_scores.clear();
for (unsigned long k = 0; k < rel_scores.size(); ++k)
{
rel_scores[k] = df(samples_test[i].relevant[k]);
total_scores.push_back(std::make_pair(rel_scores[k], true));
}
for (unsigned long k = 0; k < nonrel_scores.size(); ++k)
{
nonrel_scores[k] = df(samples_test[i].nonrelevant[k]);
total_scores.push_back(std::make_pair(nonrel_scores[k], false));
}
// Now compute the average precision for this sample. We need to sort the
// results and the back them into total_ranking. Note that we sort them so
// that, if you get a block of ranking values that are all equal, the elements
// marked as true will come last. This prevents a ranking from outputting a
// constant value for everything and still getting a good MAP score.
std::sort(total_scores.rbegin(), total_scores.rend(), impl::compare_first_reverse_second);
total_ranking.clear();
for (unsigned long i = 0; i < total_scores.size(); ++i)
total_ranking.push_back(total_scores[i].second);
rs.add(average_precision(total_ranking));
count_ranking_inversions(rel_scores, nonrel_scores, rel_counts, nonrel_counts);
total_pairs += rel_scores.size()*nonrel_scores.size();
// Note that we don't need to look at nonrel_counts since it is redundant with
// the information in rel_counts in this case.
total_wrong += sum(mat(rel_counts));
}
} // for (long i = 0; i < folds; ++i)
const double rank_swaps = static_cast<double>(total_pairs - total_wrong) / total_pairs;
const double mean_average_precision = rs.mean();
matrix<double,1,2> res;
res = rank_swaps, mean_average_precision;
return res;
}
// ----------------------------------------------------------------------------------------
}
#endif // DLIB_RANKING_ToOLS_Hh_
|