SHOGUN v0.9.0
|
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 00011 #include "regression/svr/LibSVR.h" 00012 #include "lib/io.h" 00013 00014 using namespace shogun; 00015 00016 CLibSVR::CLibSVR() 00017 : CSVM() 00018 { 00019 model=NULL; 00020 } 00021 00022 CLibSVR::CLibSVR(float64_t C, float64_t eps, CKernel* k, CLabels* lab) 00023 : CSVM() 00024 { 00025 model=NULL; 00026 00027 set_C(C,C); 00028 set_tube_epsilon(eps); 00029 set_labels(lab); 00030 set_kernel(k); 00031 } 00032 00033 CLibSVR::~CLibSVR() 00034 { 00035 free(model); 00036 } 00037 00038 bool CLibSVR::train(CFeatures* data) 00039 { 00040 ASSERT(kernel); 00041 ASSERT(labels && labels->get_num_labels()); 00042 00043 if (data) 00044 { 00045 if (labels->get_num_labels() != data->get_num_vectors()) 00046 SG_ERROR("Number of training vectors does not match number of labels\n"); 00047 kernel->init(data, data); 00048 } 00049 00050 free(model); 00051 00052 struct svm_node* x_space; 00053 00054 problem.l=labels->get_num_labels(); 00055 SG_INFO( "%d trainlabels\n", problem.l); 00056 00057 problem.y=new float64_t[problem.l]; 00058 problem.x=new struct svm_node*[problem.l]; 00059 x_space=new struct svm_node[2*problem.l]; 00060 00061 for (int32_t i=0; i<problem.l; i++) 00062 { 00063 problem.y[i]=labels->get_label(i); 00064 problem.x[i]=&x_space[2*i]; 00065 x_space[2*i].index=i; 00066 x_space[2*i+1].index=-1; 00067 } 00068 00069 int32_t weights_label[2]={-1,+1}; 00070 float64_t weights[2]={1.0,get_C2()/get_C1()}; 00071 00072 param.svm_type=EPSILON_SVR; // epsilon SVR 00073 param.kernel_type = LINEAR; 00074 param.degree = 3; 00075 param.gamma = 0; // 1/k 00076 param.coef0 = 0; 00077 param.nu = 0.5; 00078 param.kernel=kernel; 00079 param.cache_size = kernel->get_cache_size(); 00080 param.max_train_time = max_train_time; 00081 param.C = get_C1(); 00082 param.eps = epsilon; 00083 param.p = tube_epsilon; 00084 param.shrinking = 1; 00085 param.nr_weight = 2; 00086 param.weight_label = weights_label; 00087 param.weight = weights; 00088 param.use_bias = get_bias_enabled(); 00089 00090 const char* error_msg = svm_check_parameter(&problem,¶m); 00091 00092 if(error_msg) 00093 SG_ERROR("Error: %s\n",error_msg); 00094 00095 model = svm_train(&problem, ¶m); 00096 00097 if (model) 00098 { 00099 ASSERT(model->nr_class==2); 00100 ASSERT((model->l==0) || (model->l>0 && model->SV && model->sv_coef && model->sv_coef[0])); 00101 00102 int32_t num_sv=model->l; 00103 00104 create_new_model(num_sv); 00105 00106 CSVM::set_objective(model->objective); 00107 00108 set_bias(-model->rho[0]); 00109 00110 for (int32_t i=0; i<num_sv; i++) 00111 { 00112 set_support_vector(i, (model->SV[i])->index); 00113 set_alpha(i, model->sv_coef[0][i]); 00114 } 00115 00116 delete[] problem.x; 00117 delete[] problem.y; 00118 delete[] x_space; 00119 00120 svm_destroy_model(model); 00121 model=NULL; 00122 return true; 00123 } 00124 else 00125 return false; 00126 }