LDA.cpp

Go to the documentation of this file.
00001 /*
00002  * This program is free software; you can redistribute it and/or modify
00003  * it under the terms of the GNU General Public License as published by
00004  * the Free Software Foundation; either version 3 of the License, or
00005  * (at your option) any later version.
00006  *
00007  * Written (W) 1999-2008 Soeren Sonnenburg
00008  * Copyright (C) 1999-2008 Fraunhofer Institute FIRST and Max-Planck-Society
00009  */
00010 
00011 #include "lib/common.h"
00012 
00013 #ifdef HAVE_LAPACK
00014 #include "classifier/Classifier.h"
00015 #include "classifier/LinearClassifier.h"
00016 #include "classifier/LDA.h"
00017 #include "features/Labels.h"
00018 #include "lib/Mathematics.h"
00019 #include "lib/lapack.h"
00020 
00021 CLDA::CLDA(DREAL gamma)
00022 : CLinearClassifier(), m_gamma(gamma)
00023 {
00024 }
00025 
00026 CLDA::CLDA(DREAL gamma, CRealFeatures* traindat, CLabels* trainlab)
00027 : CLinearClassifier(), m_gamma(gamma)
00028 {
00029     set_features(traindat);
00030     set_labels(trainlab);
00031 }
00032 
00033 
00034 CLDA::~CLDA()
00035 {
00036 }
00037 
00038 bool CLDA::train()
00039 {
00040     ASSERT(labels);
00041     ASSERT(features);
00042     INT num_train_labels=0;
00043     INT* train_labels=labels->get_int_labels(num_train_labels);
00044     ASSERT(train_labels);
00045 
00046     INT num_feat=features->get_num_features();
00047     INT num_vec=features->get_num_vectors();
00048     ASSERT(num_vec==num_train_labels);
00049 
00050     INT* classidx_neg=new INT[num_vec];
00051     INT* classidx_pos=new INT[num_vec];
00052 
00053     INT i=0;
00054     INT j=0;
00055     INT num_neg=0;
00056     INT num_pos=0;
00057     for (i=0; i<num_train_labels; i++)
00058     {
00059         if (train_labels[i]==-1)
00060             classidx_neg[num_neg++]=i;
00061         else if (train_labels[i]==+1)
00062             classidx_pos[num_pos++]=i;
00063         else
00064         {
00065             SG_ERROR( "found label != +/- 1 bailing...");
00066             return false;
00067         }
00068     }
00069 
00070     if (num_neg<=0 && num_pos<=0)
00071     {
00072       SG_ERROR( "whooooo ? only a single class found\n");
00073         return false;
00074     }
00075 
00076     delete[] w;
00077     w=new DREAL[num_feat];
00078     w_dim=num_feat;
00079 
00080     DREAL* mean_neg=new DREAL[num_feat];
00081     memset(mean_neg,0,num_feat*sizeof(DREAL));
00082 
00083     DREAL* mean_pos=new DREAL[num_feat];
00084     memset(mean_pos,0,num_feat*sizeof(DREAL));
00085 
00086     DREAL* scatter=new DREAL[num_feat*num_feat];
00087     DREAL* buffer=new DREAL[num_feat*CMath::max(num_neg, num_pos)];
00088 
00089     //mean neg
00090     for (i=0; i<num_neg; i++)
00091     {
00092         INT vlen;
00093         bool vfree;
00094         double* vec=features->get_feature_vector(classidx_neg[i], vlen, vfree);
00095         ASSERT(vec);
00096 
00097         for (j=0; j<vlen; j++)
00098         {
00099             mean_neg[j]+=vec[j];
00100             buffer[num_feat*i+j]=vec[j];
00101         }
00102 
00103         features->free_feature_vector(vec, classidx_neg[i], vfree);
00104     }
00105 
00106     for (j=0; j<num_feat; j++)
00107         mean_neg[j]/=num_neg;
00108 
00109     for (i=0; i<num_neg; i++)
00110     {
00111         for (j=0; j<num_feat; j++)
00112             buffer[num_feat*i+j]-=mean_neg[j];
00113     }
00114     cblas_dgemm(CblasColMajor, CblasNoTrans, CblasTrans, num_feat, num_feat, num_neg, 1.0, buffer, num_feat, buffer, num_feat, 0, scatter, num_feat);
00115     
00116     //mean pos
00117     for (i=0; i<num_pos; i++)
00118     {
00119         INT vlen;
00120         bool vfree;
00121         double* vec=features->get_feature_vector(classidx_pos[i], vlen, vfree);
00122         ASSERT(vec);
00123 
00124         for (j=0; j<vlen; j++)
00125         {
00126             mean_pos[j]+=vec[j];
00127             buffer[num_feat*i+j]=vec[j];
00128         }
00129 
00130         features->free_feature_vector(vec, classidx_pos[i], vfree);
00131     }
00132 
00133     for (j=0; j<num_feat; j++)
00134         mean_pos[j]/=num_pos;
00135 
00136     for (i=0; i<num_pos; i++)
00137     {
00138         for (j=0; j<num_feat; j++)
00139             buffer[num_feat*i+j]-=mean_pos[j];
00140     }
00141     cblas_dgemm(CblasColMajor, CblasNoTrans, CblasTrans, num_feat, num_feat, num_pos, 1.0/(num_train_labels-1), buffer, num_feat, buffer, num_feat, 1.0/(num_train_labels-1), scatter, num_feat);
00142 
00143     DREAL trace=CMath::trace(scatter, num_feat, num_feat);
00144 
00145     double s=1.0-m_gamma;
00146 
00147     for (i=0; i<num_feat*num_feat; i++)
00148         scatter[i]*=s;
00149 
00150     for (i=0; i<num_feat; i++)
00151         scatter[i*num_feat+i]+= trace*m_gamma/num_feat;
00152 
00153     DREAL* inv_scatter= CMath::pinv(scatter, num_feat, num_feat, NULL);
00154 
00155     DREAL* w_pos=buffer;
00156     DREAL* w_neg=&buffer[num_feat];
00157 
00158     cblas_dsymv(CblasColMajor, CblasUpper, num_feat, 1.0, inv_scatter, num_feat, mean_pos, 1, 0, w_pos, 1);
00159     cblas_dsymv(CblasColMajor, CblasUpper, num_feat, 1.0, inv_scatter, num_feat, mean_neg, 1, 0, w_neg, 1);
00160     
00161     bias=0.5*(CMath::dot(w_neg, mean_neg, num_feat)-CMath::dot(w_pos, mean_pos, num_feat));
00162     for (i=0; i<num_feat; i++)
00163         w[i]=w_pos[i]-w_neg[i];
00164 
00165 #ifdef DEBUG_LDA
00166     SG_PRINT("bias: %f\n", bias);
00167     CMath::display_vector(w, num_feat, "w");
00168     CMath::display_vector(w_pos, num_feat, "w_pos");
00169     CMath::display_vector(w_neg, num_feat, "w_neg");
00170     CMath::display_vector(mean_pos, num_feat, "mean_pos");
00171     CMath::display_vector(mean_neg, num_feat, "mean_neg");
00172 #endif
00173 
00174     delete[] train_labels;
00175     delete[] mean_neg;
00176     delete[] mean_pos;
00177     delete[] scatter;
00178     delete[] inv_scatter;
00179     delete[] classidx_neg;
00180     delete[] classidx_pos;
00181     delete[] buffer;
00182     return true;
00183 }
00184 #endif

SHOGUN Machine Learning Toolbox - Documentation