X-Git-Url: https://www.fleuret.org/cgi-bin/gitweb/gitweb.cgi?p=folded-ctf.git;a=blobdiff_plain;f=loss_machine.cc;h=fd1d07b9915e8690b5bea02be9e5c600fe85d972;hp=63a65cc208f67348358f492adb2b1e90cb745498;hb=71a84ea2658cd96726bcf4e582010c24bf2583cf;hpb=5b78a555f6c7ff20a71d0520db63bc43e69e1f41

diff --git a/loss_machine.cc b/loss_machine.cc
index 63a65cc..fd1d07b 100644
--- a/loss_machine.cc
+++ b/loss_machine.cc
@@ -84,40 +84,6 @@ scalar_t LossMachine::loss(SampleSet *samples, scalar_t *responses) {
     }
     break;
 
-  case LOSS_EV_REGULARIZED:
-    {
-      scalar_t sum_pos = 0, sum_sq_pos = 0, nb_pos = 0, m_pos, v_pos;
-      scalar_t sum_neg = 0, sum_sq_neg = 0, nb_neg = 0, m_neg, v_neg;
-
-      for(int n = 0; n < samples->nb_samples(); n++) {
-        if(samples->label(n) > 0) {
-          sum_pos += responses[n];
-          sum_sq_pos += sq(responses[n]);
-          nb_pos += 1.0;
-        } else if(samples->label(n) < 0) {
-          sum_neg += responses[n];
-          sum_sq_neg += sq(responses[n]);
-          nb_neg += 1.0;
-        }
-      }
-
-      l = 0;
-
-      if(nb_pos > 0) {
-        m_pos = sum_pos / nb_pos;
-        v_pos = sum_sq_pos/(nb_pos - 1) - sq(sum_pos)/(nb_pos * (nb_pos - 1));
-        l += nb_pos * exp(v_pos/2 - m_pos);
-      }
-
-      if(nb_neg > 0) {
-        m_neg = sum_neg / nb_neg;
-        v_neg = sum_sq_neg/(nb_neg - 1) - sq(sum_neg)/(nb_neg * (nb_neg - 1));
-        l += nb_neg * exp(v_neg/2 + m_neg);
-      }
-
-    }
-    break;
-
   case LOSS_HINGE:
     {
       for(int n = 0; n < samples->nb_samples(); n++) {
@@ -161,6 +127,7 @@ scalar_t LossMachine::optimal_weight(SampleSet *sample_set,
   case LOSS_EXPONENTIAL:
     {
       scalar_t num = 0, den = 0, z;
+
       for(int n = 0; n < sample_set->nb_samples(); n++) {
         z = sample_set->label(n) * weak_learner_responses[n];
         if(z > 0) {
@@ -174,103 +141,6 @@ scalar_t LossMachine::optimal_weight(SampleSet *sample_set,
     }
     break;
 
-  case LOSS_EV_REGULARIZED:
-    {
-
-      scalar_t u = 0, du = -0.1;
-      scalar_t *responses = new scalar_t[sample_set->nb_samples()];
-
-      scalar_t l, prev_l = -1;
-
-      const scalar_t minimum_delta_for_optimization = 1e-5;
-
-      scalar_t shift = 0;
-
-      {
-        scalar_t sum_pos = 0, sum_sq_pos = 0, nb_pos = 0, m_pos, v_pos;
-        scalar_t sum_neg = 0, sum_sq_neg = 0, nb_neg = 0, m_neg, v_neg;
-
-        for(int n = 0; n < sample_set->nb_samples(); n++) {
-          if(sample_set->label(n) > 0) {
-            sum_pos += responses[n];
-            sum_sq_pos += sq(responses[n]);
-            nb_pos += 1.0;
-          } else if(sample_set->label(n) < 0) {
-            sum_neg += responses[n];
-            sum_sq_neg += sq(responses[n]);
-            nb_neg += 1.0;
-          }
-        }
-
-        if(nb_pos > 0) {
-          m_pos = sum_pos / nb_pos;
-          v_pos = sum_sq_pos/(nb_pos - 1) - sq(sum_pos)/(nb_pos * (nb_pos - 1));
-          shift = max(shift, v_pos/2 - m_pos);
-        }
-
-        if(nb_neg > 0) {
-          m_neg = sum_neg / nb_neg;
-          v_neg = sum_sq_neg/(nb_neg - 1) - sq(sum_neg)/(nb_neg * (nb_neg - 1));
-          shift = max(shift, v_neg/2 + m_neg);
-        }
-
-//         (*global.log_stream) << "nb_pos = " << nb_pos << " nb_neg = " << nb_neg << endl;
-
-      }
-
-      int nb = 0;
-
-      while(nb < 100 && abs(du) > minimum_delta_for_optimization) {
-        nb++;
-
-//         (*global.log_stream) << "l = " << l << " u = " << u << " du = " << du << endl;
-
-        u += du;
-        for(int s = 0; s < sample_set->nb_samples(); s++) {
-          responses[s] = current_responses[s] + u * weak_learner_responses[s] ;
-        }
-
-        {
-          scalar_t sum_pos = 0, sum_sq_pos = 0, nb_pos = 0, m_pos, v_pos;
-          scalar_t sum_neg = 0, sum_sq_neg = 0, nb_neg = 0, m_neg, v_neg;
-
-          for(int n = 0; n < sample_set->nb_samples(); n++) {
-            if(sample_set->label(n) > 0) {
-              sum_pos += responses[n];
-              sum_sq_pos += sq(responses[n]);
-              nb_pos += 1.0;
-            } else if(sample_set->label(n) < 0) {
-              sum_neg += responses[n];
-              sum_sq_neg += sq(responses[n]);
-              nb_neg += 1.0;
-            }
-          }
-
-          l = 0;
-
-          if(nb_pos > 0) {
-            m_pos = sum_pos / nb_pos;
-            v_pos = sum_sq_pos/(nb_pos - 1) - sq(sum_pos)/(nb_pos * (nb_pos - 1));
-            l += nb_pos * exp(v_pos/2 - m_pos - shift);
-          }
-
-          if(nb_neg > 0) {
-            m_neg = sum_neg / nb_neg;
-            v_neg = sum_sq_neg/(nb_neg - 1) - sq(sum_neg)/(nb_neg * (nb_neg - 1));
-            l += nb_neg * exp(v_neg/2 + m_neg - shift);
-          }
-
-        }
-
-        if(l > prev_l) du = du * -0.25;
-        prev_l = l;
-      }
-
-      delete[] responses;
-
-      return u;
-    }
-
   case LOSS_HINGE:
   case LOSS_LOGISTIC:
     {
@@ -349,8 +219,6 @@ void LossMachine::subsample(int nb, scalar_t *labels, scalar_t *responses,
         }
       } while(nb_sampled < nb_to_sample);
 
-      (*global.log_stream) << "nb_sampled = " << nb_sampled << " nb_to_sample = " << nb_to_sample << endl;
-
       (*global.log_stream) << "Done." << endl;
 
       delete[] sampled_indexes;