LibLinear.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 #include "lib/config.h"
00011 
00012 #ifdef HAVE_LAPACK
00013 #include "lib/io.h"
00014 #include "classifier/svm/LibLinear.h"
00015 #include "classifier/svm/SVM_linear.h"
00016 #include "classifier/svm/Tron.h"
00017 #include "features/SparseFeatures.h"
00018 
00019 CLibLinear::CLibLinear(LIBLINEAR_LOSS l)
00020 : CSparseLinearClassifier()
00021 {
00022     loss=l;
00023     use_bias=false;
00024     C1=1;
00025     C2=1;
00026 }
00027 
00028 CLibLinear::CLibLinear(
00029     float64_t C, CSparseFeatures<float64_t>* traindat, CLabels* trainlab)
00030 : CSparseLinearClassifier(), C1(C), C2(C), use_bias(true), epsilon(1e-5)
00031 {
00032     set_features(traindat);
00033     set_labels(trainlab);
00034     loss=LR;
00035 }
00036 
00037 
00038 CLibLinear::~CLibLinear()
00039 {
00040 }
00041 
00042 bool CLibLinear::train()
00043 {
00044     ASSERT(labels);
00045     ASSERT(get_features());
00046     ASSERT(labels->is_two_class_labeling());
00047 
00048     CSparseFeatures<float64_t>* sfeat=(CSparseFeatures<float64_t>*) features;
00049 
00050     int32_t num_train_labels=labels->get_num_labels();
00051     int32_t num_feat=features->get_num_features();
00052     int32_t num_vec=features->get_num_vectors();
00053 
00054     ASSERT(num_vec==num_train_labels);
00055     delete[] w;
00056     if (use_bias)
00057         w=new float64_t[num_feat+1];
00058     else
00059         w=new float64_t[num_feat+0];
00060     w_dim=num_feat;
00061 
00062     problem prob;
00063     if (use_bias)
00064     {
00065         prob.n=w_dim+1;
00066         memset(w, 0, sizeof(float64_t)*(w_dim+1));
00067     }
00068     else
00069     {
00070         prob.n=w_dim;
00071         memset(w, 0, sizeof(float64_t)*(w_dim+0));
00072     }
00073     prob.l=num_vec;
00074     prob.x=sfeat;
00075     prob.y=new int[prob.l];
00076     prob.use_bias=use_bias;
00077 
00078     for (int32_t i=0; i<prob.l; i++)
00079         prob.y[i]=labels->get_int_label(i);
00080 
00081     SG_INFO( "%d training points %d dims\n", prob.l, prob.n);
00082 
00083     function *fun_obj=NULL;
00084 
00085     switch (loss)
00086     {
00087         case LR:
00088             fun_obj=new l2_lr_fun(&prob, get_C1(), get_C2());
00089             break;
00090         case L2:
00091             fun_obj=new l2loss_svm_fun(&prob, get_C1(), get_C2());
00092             break;
00093         default:
00094             SG_ERROR("unknown loss\n");
00095             break;
00096     }
00097 
00098     if (fun_obj)
00099     {
00100         CTron tron_obj(fun_obj, epsilon);
00101         tron_obj.tron(w);
00102         float64_t sgn=prob.y[0];
00103 
00104         for (int32_t i=0; i<w_dim; i++)
00105             w[i]*=sgn;
00106 
00107         if (use_bias)
00108             set_bias(sgn*w[w_dim]);
00109         else
00110             set_bias(0);
00111 
00112         delete fun_obj;
00113     }
00114 
00115     delete[] prob.y;
00116 
00117     return true;
00118 }
00119 #endif //HAVE_LAPACK

SHOGUN Machine Learning Toolbox - Documentation