SHOGUN
v1.1.0
|
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) 2011 Hidekazu Oiwa 00008 */ 00009 00010 #include <shogun/classifier/AveragedPerceptron.h> 00011 #include <shogun/features/Labels.h> 00012 #include <shogun/mathematics/Math.h> 00013 00014 using namespace shogun; 00015 00016 CAveragedPerceptron::CAveragedPerceptron() 00017 : CLinearMachine(), learn_rate(0.1), max_iter(1000) 00018 { 00019 } 00020 00021 CAveragedPerceptron::CAveragedPerceptron(CDotFeatures* traindat, CLabels* trainlab) 00022 : CLinearMachine(), learn_rate(.1), max_iter(1000) 00023 { 00024 set_features(traindat); 00025 set_labels(trainlab); 00026 } 00027 00028 CAveragedPerceptron::~CAveragedPerceptron() 00029 { 00030 } 00031 00032 bool CAveragedPerceptron::train(CFeatures* data) 00033 { 00034 ASSERT(labels); 00035 if (data) 00036 { 00037 if (!data->has_property(FP_DOT)) 00038 SG_ERROR("Specified features are not of type CDotFeatures\n"); 00039 set_features((CDotFeatures*) data); 00040 } 00041 ASSERT(features); 00042 bool converged=false; 00043 int32_t iter=0; 00044 SGVector<int32_t> train_labels=labels->get_int_labels(); 00045 int32_t num_feat=features->get_dim_feature_space(); 00046 int32_t num_vec=features->get_num_vectors(); 00047 00048 ASSERT(num_vec==train_labels.vlen); 00049 SG_FREE(w); 00050 w_dim=num_feat; 00051 w=SG_MALLOC(float64_t, num_feat); 00052 float64_t* tmp_w=SG_MALLOC(float64_t, num_feat); 00053 00054 float64_t* output=SG_MALLOC(float64_t, num_vec); 00055 //start with uniform w, bias=0, tmp_bias=0 00056 bias=0; 00057 float64_t tmp_bias=0; 00058 for (int32_t i=0; i<num_feat; i++) 00059 w[i]=1.0/num_feat; 00060 00061 //loop till we either get everything classified right or reach max_iter 00062 00063 while (!converged && iter<max_iter) 00064 { 00065 converged=true; 00066 for (int32_t i=0; i<num_vec; i++) 00067 { 00068 output[i]=apply(i); 00069 00070 if (CMath::sign<float64_t>(output[i]) != train_labels.vector[i]) 00071 { 00072 converged=false; 00073 bias+=learn_rate*train_labels.vector[i]; 00074 features->add_to_dense_vec(learn_rate*train_labels.vector[i], i, w, w_dim); 00075 } 00076 00077 // Add current w to tmp_w, and current bias to tmp_bias 00078 // To calculate the sum of each iteration's w, bias 00079 for (int32_t j=0; j<num_feat; j++) 00080 tmp_w[j]+=w[j]; 00081 tmp_bias+=bias; 00082 } 00083 iter++; 00084 } 00085 00086 if (converged) 00087 SG_INFO("Averaged Perceptron algorithm converged after %d iterations.\n", iter); 00088 else 00089 SG_WARNING("Averaged Perceptron algorithm did not converge after %d iterations.\n", max_iter); 00090 00091 // calculate and set the average paramter of w, bias 00092 for (int32_t i=0; i<num_feat; i++) 00093 w[i]=tmp_w[i]/(num_vec*iter); 00094 bias=tmp_bias/(num_vec*iter); 00095 00096 SG_FREE(output); 00097 train_labels.free_vector(); 00098 SG_FREE(tmp_w); 00099 00100 return converged; 00101 }