2 ///////////////////////////////////////////////////////////////////////////
3 // This program is free software: you can redistribute it and/or modify //
4 // it under the terms of the version 3 of the GNU General Public License //
5 // as published by the Free Software Foundation. //
7 // This program is distributed in the hope that it will be useful, but //
8 // WITHOUT ANY WARRANTY; without even the implied warranty of //
9 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU //
10 // General Public License for more details. //
12 // You should have received a copy of the GNU General Public License //
13 // along with this program. If not, see <http://www.gnu.org/licenses/>. //
15 // Written by Francois Fleuret //
16 // (C) Idiap Research Institute //
18 // Contact <francois.fleuret@idiap.ch> for comments & bug reports //
19 ///////////////////////////////////////////////////////////////////////////
22 #include "loss_machine.h"
24 LossMachine::LossMachine(int loss_type) {
25 _loss_type = loss_type;
28 void LossMachine::get_loss_derivatives(SampleSet *samples,
30 scalar_t *derivatives) {
34 case LOSS_EXPONENTIAL:
36 for(int n = 0; n < samples->nb_samples(); n++) {
38 - samples->label(n) * exp( - samples->label(n) * responses[n]);
45 for(int n = 0; n < samples->nb_samples(); n++) {
46 if(samples->label(n) != 0 && samples->label(n) * responses[n] < 1)
56 for(int n = 0; n < samples->nb_samples(); n++) {
57 if(samples->label(n) == 0)
60 derivatives[n] = samples->label(n) * 1/(1 + exp(samples->label(n) * responses[n]));
66 cerr << "Unknown loss type in BoostedClassifier::get_loss_derivatives."
73 scalar_t LossMachine::loss(SampleSet *samples, scalar_t *responses) {
78 case LOSS_EXPONENTIAL:
80 for(int n = 0; n < samples->nb_samples(); n++) {
81 l += exp( - samples->label(n) * responses[n]);
89 for(int n = 0; n < samples->nb_samples(); n++) {
90 if(samples->label(n) != 0) {
91 if(samples->label(n) * responses[n] < 1)
92 l += (1 - samples->label(n) * responses[n]);
100 for(int n = 0; n < samples->nb_samples(); n++) {
101 if(samples->label(n) != 0) {
102 scalar_t u = - samples->label(n) * responses[n];
106 l += log(1 + exp(u));
114 cerr << "Unknown loss type in LossMachine::loss." << endl;
121 scalar_t LossMachine::optimal_weight(SampleSet *sample_set,
122 scalar_t *weak_learner_responses,
123 scalar_t *current_responses) {
127 case LOSS_EXPONENTIAL:
129 scalar_t num = 0, den = 0, z;
131 for(int n = 0; n < sample_set->nb_samples(); n++) {
132 z = sample_set->label(n) * weak_learner_responses[n];
134 num += exp( - sample_set->label(n) * current_responses[n]);
136 den += exp( - sample_set->label(n) * current_responses[n]);
140 return 0.5 * log(num / den);
148 scalar_t u = 0, du = -0.1;
149 scalar_t *responses = new scalar_t[sample_set->nb_samples()];
151 scalar_t l, prev_l = -1;
153 const scalar_t minimum_delta_for_optimization = 1e-5;
156 while(n < 100 && abs(du) > minimum_delta_for_optimization) {
159 for(int s = 0; s < sample_set->nb_samples(); s++) {
160 responses[s] = current_responses[s] + u * weak_learner_responses[s] ;
162 l = loss(sample_set, responses);
163 if(l > prev_l) du = du * -0.25;
167 (*global.log_stream) << "END l = " << l << " du = " << du << endl;
175 cerr << "Unknown loss type in LossMachine::optimal_weight." << endl;
181 void LossMachine::subsample(int nb, scalar_t *labels, scalar_t *responses,
182 int nb_to_sample, int *sample_nb_occurences, scalar_t *sample_responses,
183 int allow_duplicates) {
187 case LOSS_EXPONENTIAL:
189 scalar_t *weights = new scalar_t[nb];
191 for(int n = 0; n < nb; n++) {
195 weights[n] = exp( - labels[n] * responses[n]);
197 sample_nb_occurences[n] = 0;
198 sample_responses[n] = 0.0;
201 scalar_t total_weight;
202 int nb_sampled = 0, sum_sample_nb_occurences = 0;
204 int *sampled_indexes = new int[nb_to_sample];
206 (*global.log_stream) << "Sampling " << nb_to_sample << " samples." << endl;
209 total_weight = robust_sampling(nb,
214 for(int k = 0; nb_sampled < nb_to_sample && k < nb_to_sample; k++) {
215 int i = sampled_indexes[k];
216 if(allow_duplicates || sample_nb_occurences[i] == 0) nb_sampled++;
217 sample_nb_occurences[i]++;
218 sum_sample_nb_occurences++;
220 } while(nb_sampled < nb_to_sample);
222 (*global.log_stream) << "Done." << endl;
224 delete[] sampled_indexes;
226 scalar_t unit_weight = log(total_weight / scalar_t(sum_sample_nb_occurences));
228 for(int n = 0; n < nb; n++) {
229 if(sample_nb_occurences[n] > 0) {
230 if(allow_duplicates) {
231 sample_responses[n] = - labels[n] * unit_weight;
233 sample_responses[n] = - labels[n] * (unit_weight + log(scalar_t(sample_nb_occurences[n])));
234 sample_nb_occurences[n] = 1;
245 cerr << "Unknown loss type in LossMachine::resample." << endl;