/usr/include/vigra/random_forest/rf_earlystopping.hxx is in libvigraimpex-dev 1.10.0+dfsg-3ubuntu2.
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 | #ifndef RF_EARLY_STOPPING_P_HXX
#define RF_EARLY_STOPPING_P_HXX
#include <cmath>
#include "rf_common.hxx"
namespace vigra
{
#if 0
namespace es_detail
{
template<class T>
T power(T const & in, int n)
{
T result = NumericTraits<T>::one();
for(int ii = 0; ii < n ;++ii)
result *= in;
return result;
}
}
#endif
/**Base class from which all EarlyStopping Functors derive.
*/
class StopBase
{
protected:
ProblemSpec<> ext_param_;
int tree_count_ ;
bool is_weighted_;
public:
template<class T>
void set_external_parameters(ProblemSpec<T> const &prob, int tree_count = 0, bool is_weighted = false)
{
ext_param_ = prob;
is_weighted_ = is_weighted;
tree_count_ = tree_count;
}
#ifdef DOXYGEN
/** called after the prediction of a tree was added to the total prediction
* \param weightIter Iterator to the weights delivered by current tree.
* \param k after kth tree
* \param prob Total probability array
* \param totalCt sum of probability array.
*/
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter weightIter, int k, MultiArrayView<2, T, C> const & prob , double totalCt)
#else
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter, int /* k */, MultiArrayView<2, T, C> const & /* prob */, double /* totalCt */)
{return false;}
#endif //DOXYGEN
};
/**Stop predicting after a set number of trees
*/
class StopAfterTree : public StopBase
{
public:
double max_tree_p;
int max_tree_;
typedef StopBase SB;
ArrayVector<double> depths;
/** Constructor
* \param max_tree number of trees to be used for prediction
*/
StopAfterTree(double max_tree)
:
max_tree_p(max_tree)
{}
template<class T>
void set_external_parameters(ProblemSpec<T> const &prob, int tree_count = 0, bool is_weighted = false)
{
max_tree_ = ceil(max_tree_p * tree_count);
SB::set_external_parameters(prob, tree_count, is_weighted);
}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter, int k, MultiArrayView<2, T, C> const & /* prob */, double /* totalCt */)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
if(k < max_tree_)
return false;
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
};
/** Stop predicting after a certain amount of votes exceed certain proportion.
* case unweighted voting: stop if the leading class exceeds proportion * SB::tree_count_
* case weighted voting: stop if the leading class exceeds proportion * msample_ * SB::tree_count_ ;
* (maximal number of votes possible in both cases)
*/
class StopAfterVoteCount : public StopBase
{
public:
double proportion_;
typedef StopBase SB;
ArrayVector<double> depths;
/** Constructor
* \param proportion specify proportion to be used.
*/
StopAfterVoteCount(double proportion)
:
proportion_(proportion)
{}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter, int k, MultiArrayView<2, T, C> const & prob, double /* totalCt */)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
if(SB::is_weighted_)
{
if(prob[argMax(prob)] > proportion_ *SB::ext_param_.actual_msample_* SB::tree_count_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
}
else
{
if(prob[argMax(prob)] > proportion_ * SB::tree_count_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
}
return false;
}
};
/** Stop predicting if the 2norm of the probabilities does not change*/
class StopIfConverging : public StopBase
{
public:
double thresh_;
int num_;
MultiArray<2, double> last_;
MultiArray<2, double> cur_;
ArrayVector<double> depths;
typedef StopBase SB;
/** Constructor
* \param thresh: If the two norm of the probabilities changes less then thresh then stop
* \param num : look at atleast num trees before stopping
*/
StopIfConverging(double thresh, int num = 10)
:
thresh_(thresh),
num_(num)
{}
template<class T>
void set_external_parameters(ProblemSpec<T> const &prob, int tree_count = 0, bool is_weighted = false)
{
last_.reshape(MultiArrayShape<2>::type(1, prob.class_count_), 0);
cur_.reshape(MultiArrayShape<2>::type(1, prob.class_count_), 0);
SB::set_external_parameters(prob, tree_count, is_weighted);
}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter iter, int k, MultiArrayView<2, T, C> const & prob, double totalCt)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
if(k <= num_)
{
last_ = prob;
last_/= last_.norm(1);
return false;
}
else
{
cur_ = prob;
cur_ /= cur_.norm(1);
last_ -= cur_;
double nrm = last_.norm();
if(nrm < thresh_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
else
{
last_ = cur_;
}
}
return false;
}
};
/** Stop predicting if the margin prob(leading class) - prob(second class) exceeds a proportion
* case unweighted voting: stop if margin exceeds proportion * SB::tree_count_
* case weighted voting: stop if margin exceeds proportion * msample_ * SB::tree_count_ ;
* (maximal number of votes possible in both cases)
*/
class StopIfMargin : public StopBase
{
public:
double proportion_;
typedef StopBase SB;
ArrayVector<double> depths;
/** Constructor
* \param proportion specify proportion to be used.
*/
StopIfMargin(double proportion)
:
proportion_(proportion)
{}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter, int k, MultiArrayView<2, T, C> prob, double /* totalCt */)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
int index = argMax(prob);
double a = prob[argMax(prob)];
prob[argMax(prob)] = 0;
double b = prob[argMax(prob)];
prob[index] = a;
double margin = a - b;
if(SB::is_weighted_)
{
if(margin > proportion_ *SB::ext_param_.actual_msample_ * SB::tree_count_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
}
else
{
if(prob[argMax(prob)] > proportion_ * SB::tree_count_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
}
return false;
}
};
/**Probabilistic Stopping criterion (binomial test)
*
* Can only be used in a two class setting
*
* Stop if the Parameters estimated for the underlying binomial distribution
* can be estimated with certainty over 1-alpha.
* (Thesis, Rahul Nair Page 80 onwards: called the "binomial" criterion
*/
class StopIfBinTest : public StopBase
{
public:
double alpha_;
MultiArrayView<2, double> n_choose_k;
/** Constructor
* \param alpha specify alpha (=proportion) value for binomial test.
* \param nck_ Matrix with precomputed values for n choose k
* nck_(n, k) is n choose k.
*/
StopIfBinTest(double alpha, MultiArrayView<2, double> nck_)
:
alpha_(alpha),
n_choose_k(nck_)
{}
typedef StopBase SB;
/**ArrayVector that will contain the fraction of trees that was visited before terminating
*/
ArrayVector<double> depths;
double binomial(int N, int k, double p)
{
// return n_choose_k(N, k) * es_detail::power(p, k) *es_detail::power(1 - p, N-k);
return n_choose_k(N, k) * std::pow(p, k) * std::pow(1 - p, N-k);
}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter iter, int k, MultiArrayView<2, T, C> prob, double totalCt)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
if(k < 10)
{
return false;
}
int index = argMax(prob);
int n_a = prob[index];
int n_b = prob[(index+1)%2];
int n_tilde = (SB::tree_count_ - n_a + n_b);
double p_a = double(n_b - n_a + n_tilde)/double(2* n_tilde);
vigra_precondition(p_a <= 1, "probability should be smaller than 1");
double cum_val = 0;
int c = 0;
// std::cerr << "prob: " << p_a << std::endl;
if(n_a <= 0)n_a = 0;
if(n_b <= 0)n_b = 0;
for(int ii = 0; ii <= n_b + n_a;++ii)
{
// std::cerr << "nb +ba " << n_b + n_a << " " << ii <<std::endl;
cum_val += binomial(n_b + n_a, ii, p_a);
if(cum_val >= 1 -alpha_)
{
c = ii;
break;
}
}
// std::cerr << c << " " << n_a << " " << n_b << " " << p_a << alpha_ << std::endl;
if(c < n_a)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
return false;
}
};
/**Probabilistic Stopping criteria. (toChange)
*
* Can only be used in a two class setting
*
* Stop if the probability that the decision will change after seeing all trees falls under
* a specified value alpha.
* (Thesis, Rahul Nair Page 80 onwards: called the "toChange" criterion
*/
class StopIfProb : public StopBase
{
public:
double alpha_;
MultiArrayView<2, double> n_choose_k;
/** Constructor
* \param alpha specify alpha (=proportion) value
* \param nck_ Matrix with precomputed values for n choose k
* nck_(n, k) is n choose k.
*/
StopIfProb(double alpha, MultiArrayView<2, double> nck_)
:
alpha_(alpha),
n_choose_k(nck_)
{}
typedef StopBase SB;
/**ArrayVector that will contain the fraction of trees that was visited before terminating
*/
ArrayVector<double> depths;
double binomial(int N, int k, double p)
{
// return n_choose_k(N, k) * es_detail::power(p, k) *es_detail::power(1 - p, N-k);
return n_choose_k(N, k) * std::pow(p, k) * std::pow(1 - p, N-k);
}
template<class WeightIter, class T, class C>
bool after_prediction(WeightIter iter, int k, MultiArrayView<2, T, C> prob, double totalCt)
{
if(k == SB::tree_count_ -1)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return false;
}
if(k <= 10)
{
return false;
}
int index = argMax(prob);
int n_a = prob[index];
int n_b = prob[(index+1)%2];
int n_needed = ceil(double(SB::tree_count_)/2.0)-n_a;
int n_tilde = SB::tree_count_ - (n_a +n_b);
if(n_tilde <= 0) n_tilde = 0;
if(n_needed <= 0) n_needed = 0;
double p = 0;
for(int ii = n_needed; ii < n_tilde; ++ii)
p += binomial(n_tilde, ii, 0.5);
if(p >= 1-alpha_)
{
depths.push_back(double(k+1)/double(SB::tree_count_));
return true;
}
return false;
}
};
} //namespace vigra;
#endif //RF_EARLY_STOPPING_P_HXX
|