WDSVMOcas.h

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) 2007-2008 Vojtech Franc
00008  * Written (W) 2007-2009 Soeren Sonnenburg
00009  * Copyright (C) 2007-2009 Fraunhofer Institute FIRST and Max-Planck-Society
00010  */
00011 
00012 #ifndef _WDSVMOCAS_H___
00013 #define _WDSVMOCAS_H___
00014 
00015 #include "lib/common.h"
00016 #include "classifier/Classifier.h"
00017 #include "classifier/svm/SVMOcas.h"
00018 #include "features/StringFeatures.h"
00019 #include "features/Labels.h"
00020 
00021 namespace shogun
00022 {
00023 template <class ST> class CStringFeatures;
00024 
00026 class CWDSVMOcas : public CClassifier
00027 {
00028     public:
00033         CWDSVMOcas(E_SVM_TYPE type);
00034 
00043         CWDSVMOcas(
00044             float64_t C, int32_t d, int32_t from_d,
00045             CStringFeatures<uint8_t>* traindat, CLabels* trainlab);
00046         virtual ~CWDSVMOcas();
00047 
00052         virtual inline EClassifierType get_classifier_type() { return CT_WDSVMOCAS; }
00053 
00062         virtual bool train(CFeatures* data=NULL);
00063 
00069         inline void set_C(float64_t c1, float64_t c2) { C1=c1; C2=c2; }
00070 
00075         inline float64_t get_C1() { return C1; }
00076 
00081         inline float64_t get_C2() { return C2; }
00082 
00087         inline void set_epsilon(float64_t eps) { epsilon=eps; }
00088 
00093         inline float64_t get_epsilon() { return epsilon; }
00094 
00099         inline void set_features(CStringFeatures<uint8_t>* feat)
00100         {
00101             SG_UNREF(features);
00102             SG_REF(feat);
00103             features=feat;
00104         }
00105 
00110         inline CStringFeatures<uint8_t>* get_features()
00111         {
00112             SG_REF(features);
00113             return features;
00114         }
00115 
00120         inline void set_bias_enabled(bool enable_bias) { use_bias=enable_bias; }
00121 
00126         inline bool get_bias_enabled() { return use_bias; }
00127 
00132         inline void set_bufsize(int32_t sz) { bufsize=sz; }
00133 
00138         inline int32_t get_bufsize() { return bufsize; }
00139 
00145         inline void set_degree(int32_t d, int32_t from_d)
00146         {
00147             degree=d;
00148             from_degree=from_d;
00149         }
00150 
00155         inline int32_t get_degree() { return degree; }
00156 
00161         CLabels* classify();
00162 
00168         virtual CLabels* classify(CFeatures* data);
00169 
00175         inline virtual float64_t classify_example(int32_t num)
00176         {
00177             ASSERT(features);
00178             if (!wd_weights)
00179                 set_wd_weights();
00180 
00181             int32_t len=0;
00182             float64_t sum=0;
00183             bool free_vec;
00184             uint8_t* vec=features->get_feature_vector(num, len, free_vec);
00185             //SG_INFO("len %d, string_length %d\n", len, string_length);
00186             ASSERT(len==string_length);
00187 
00188             for (int32_t j=0; j<string_length; j++)
00189             {
00190                 int32_t offs=w_dim_single_char*j;
00191                 int32_t val=0;
00192                 for (int32_t k=0; (j+k<string_length) && (k<degree); k++)
00193                 {
00194                     val=val*alphabet_size + vec[j+k];
00195                     sum+=wd_weights[k] * w[offs+val];
00196                     offs+=w_offsets[k];
00197                 }
00198             }
00199             features->free_feature_vector(vec, len, free_vec);
00200             return sum/normalization_const;
00201         }
00202 
00204         inline void set_normalization_const()
00205         {
00206             ASSERT(features);
00207             normalization_const=0;
00208             for (int32_t i=0; i<degree; i++)
00209                 normalization_const+=(string_length-i)*wd_weights[i]*wd_weights[i];
00210 
00211             normalization_const=CMath::sqrt(normalization_const);
00212             SG_DEBUG("normalization_const:%f\n", normalization_const);
00213         }
00214 
00219         inline float64_t get_normalization_const() { return normalization_const; }
00220 
00221 
00222     protected:
00227         int32_t set_wd_weights();
00228 
00237         static void compute_W(
00238             float64_t *sq_norm_W, float64_t *dp_WoldW, float64_t *alpha,
00239             uint32_t nSel, void* ptr );
00240 
00247         static float64_t update_W(float64_t t, void* ptr );
00248 
00254         static void* add_new_cut_helper(void* ptr);
00255 
00264         static void add_new_cut(
00265             float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length,
00266             uint32_t nSel, void* ptr );
00267 
00273         static void* compute_output_helper(void* ptr);
00274 
00280         static void compute_output( float64_t *output, void* ptr );
00281 
00288         static void sort( float64_t* vals, uint32_t* idx, uint32_t size);
00289 
00291         inline virtual const char* get_name() const { return "WDSVMOcas"; }
00292 
00293     protected:
00295         CStringFeatures<uint8_t>* features;
00297         bool use_bias;
00299         int32_t bufsize;
00301         float64_t C1;
00303         float64_t C2;
00305         float64_t epsilon;
00307         E_SVM_TYPE method;
00308 
00310         int32_t degree;
00312         int32_t from_degree;
00314         float32_t* wd_weights;
00316         int32_t num_vec;
00318         int32_t string_length;
00320         int32_t alphabet_size;
00321 
00323         float64_t normalization_const;
00324 
00326         float64_t bias;
00328         float64_t old_bias;
00330         int32_t* w_offsets;
00332         int32_t w_dim;
00334         int32_t w_dim_single_char;
00336         float32_t* w;
00338         float32_t* old_w;
00340         float64_t* lab;
00341 
00343         float32_t** cuts;
00345         float64_t* cp_bias;
00346 };
00347 }
00348 #endif

SHOGUN Machine Learning Toolbox - Documentation