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) 1999-2009 Soeren Sonnenburg
00008  * Copyright (C) 1999-2009 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/DotFeatures.h"
00018 
00019 using namespace shogun;
00020 
00021 CLibLinear::CLibLinear(LIBLINEAR_LOSS l)
00022 : CLinearClassifier()
00023 {
00024     loss=l;
00025     use_bias=false;
00026     C1=1;
00027     C2=1;
00028 }
00029 
00030 CLibLinear::CLibLinear(
00031     float64_t C, CDotFeatures* traindat, CLabels* trainlab)
00032 : CLinearClassifier(), C1(C), C2(C), use_bias(true), epsilon(1e-5)
00033 {
00034     set_features(traindat);
00035     set_labels(trainlab);
00036     loss=LR;
00037 }
00038 
00039 
00040 CLibLinear::~CLibLinear()
00041 {
00042 }
00043 
00044 bool CLibLinear::train(CFeatures* data)
00045 {
00046     ASSERT(labels);
00047     if (data)
00048     {
00049         if (!data->has_property(FP_DOT))
00050             SG_ERROR("Specified features are not of type CDotFeatures\n");
00051 
00052         set_features((CDotFeatures*) data);
00053     }
00054     ASSERT(features);
00055     ASSERT(labels->is_two_class_labeling());
00056 
00057     int32_t num_train_labels=labels->get_num_labels();
00058     int32_t num_feat=features->get_dim_feature_space();
00059     int32_t num_vec=features->get_num_vectors();
00060 
00061     ASSERT(num_vec==num_train_labels);
00062     delete[] w;
00063     if (use_bias)
00064         w=new float64_t[num_feat+1];
00065     else
00066         w=new float64_t[num_feat+0];
00067     w_dim=num_feat;
00068 
00069     problem prob;
00070     if (use_bias)
00071     {
00072         prob.n=w_dim+1;
00073         memset(w, 0, sizeof(float64_t)*(w_dim+1));
00074     }
00075     else
00076     {
00077         prob.n=w_dim;
00078         memset(w, 0, sizeof(float64_t)*(w_dim+0));
00079     }
00080     prob.l=num_vec;
00081     prob.x=features;
00082     prob.y=new int[prob.l];
00083     prob.use_bias=use_bias;
00084 
00085     for (int32_t i=0; i<prob.l; i++)
00086         prob.y[i]=labels->get_int_label(i);
00087 
00088     SG_INFO( "%d training points %d dims\n", prob.l, prob.n);
00089 
00090     function *fun_obj=NULL;
00091 
00092     switch (loss)
00093     {
00094         case LR:
00095             fun_obj=new l2_lr_fun(&prob, get_C1(), get_C2());
00096             break;
00097         case L2:
00098             fun_obj=new l2loss_svm_fun(&prob, get_C1(), get_C2());
00099             break;
00100         default:
00101             SG_ERROR("unknown loss\n");
00102             break;
00103     }
00104 
00105     if (fun_obj)
00106     {
00107         CTron tron_obj(fun_obj, epsilon);
00108         tron_obj.tron(w);
00109         float64_t sgn=prob.y[0];
00110 
00111         for (int32_t i=0; i<w_dim; i++)
00112             w[i]*=sgn;
00113 
00114         if (use_bias)
00115             set_bias(sgn*w[w_dim]);
00116         else
00117             set_bias(0);
00118 
00119         delete fun_obj;
00120     }
00121 
00122     delete[] prob.y;
00123 
00124     return true;
00125 }
00126 #endif //HAVE_LAPACK

SHOGUN Machine Learning Toolbox - Documentation