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 <shogun/lib/common.h>
00016 #include <shogun/machine/Machine.h>
00017 #include <shogun/classifier/svm/SVMOcas.h>
00018 #include <shogun/features/StringFeatures.h>
00019 #include <shogun/features/Labels.h>
00020 
00021 namespace shogun
00022 {
00023 template <class ST> class CStringFeatures;
00024 
00026 class CWDSVMOcas : public CMachine
00027 {
00028     public:
00030         CWDSVMOcas();
00031 
00036         CWDSVMOcas(E_SVM_TYPE type);
00037 
00046         CWDSVMOcas(
00047             float64_t C, int32_t d, int32_t from_d,
00048             CStringFeatures<uint8_t>* traindat, CLabels* trainlab);
00049         virtual ~CWDSVMOcas();
00050 
00055         virtual inline EClassifierType get_classifier_type() { return CT_WDSVMOCAS; }
00056 
00063         inline void set_C(float64_t c_neg, float64_t c_pos) { C1=c_neg; C2=c_pos; }
00064 
00069         inline float64_t get_C1() { return C1; }
00070 
00075         inline float64_t get_C2() { return C2; }
00076 
00081         inline void set_epsilon(float64_t eps) { epsilon=eps; }
00082 
00087         inline float64_t get_epsilon() { return epsilon; }
00088 
00093         inline void set_features(CStringFeatures<uint8_t>* feat)
00094         {
00095             SG_UNREF(features);
00096             SG_REF(feat);
00097             features=feat;
00098         }
00099 
00104         inline CStringFeatures<uint8_t>* get_features()
00105         {
00106             SG_REF(features);
00107             return features;
00108         }
00109 
00114         inline void set_bias_enabled(bool enable_bias) { use_bias=enable_bias; }
00115 
00120         inline bool get_bias_enabled() { return use_bias; }
00121 
00126         inline void set_bufsize(int32_t sz) { bufsize=sz; }
00127 
00132         inline int32_t get_bufsize() { return bufsize; }
00133 
00139         inline void set_degree(int32_t d, int32_t from_d)
00140         {
00141             degree=d;
00142             from_degree=from_d;
00143         }
00144 
00149         inline int32_t get_degree() { return degree; }
00150 
00155         CLabels* apply();
00156 
00162         virtual CLabels* apply(CFeatures* data);
00163 
00169         inline virtual float64_t apply(int32_t num)
00170         {
00171             ASSERT(features);
00172             if (!wd_weights)
00173                 set_wd_weights();
00174 
00175             int32_t len=0;
00176             float64_t sum=0;
00177             bool free_vec;
00178             uint8_t* vec=features->get_feature_vector(num, len, free_vec);
00179             //SG_INFO("len %d, string_length %d\n", len, string_length);
00180             ASSERT(len==string_length);
00181 
00182             for (int32_t j=0; j<string_length; j++)
00183             {
00184                 int32_t offs=w_dim_single_char*j;
00185                 int32_t val=0;
00186                 for (int32_t k=0; (j+k<string_length) && (k<degree); k++)
00187                 {
00188                     val=val*alphabet_size + vec[j+k];
00189                     sum+=wd_weights[k] * w[offs+val];
00190                     offs+=w_offsets[k];
00191                 }
00192             }
00193             features->free_feature_vector(vec, num, free_vec);
00194             return sum/normalization_const;
00195         }
00196 
00198         inline void set_normalization_const()
00199         {
00200             ASSERT(features);
00201             normalization_const=0;
00202             for (int32_t i=0; i<degree; i++)
00203                 normalization_const+=(string_length-i)*wd_weights[i]*wd_weights[i];
00204 
00205             normalization_const=CMath::sqrt(normalization_const);
00206             SG_DEBUG("normalization_const:%f\n", normalization_const);
00207         }
00208 
00213         inline float64_t get_normalization_const() { return normalization_const; }
00214 
00215 
00216     protected:
00221         int32_t set_wd_weights();
00222 
00231         static void compute_W(
00232             float64_t *sq_norm_W, float64_t *dp_WoldW, float64_t *alpha,
00233             uint32_t nSel, void* ptr );
00234 
00241         static float64_t update_W(float64_t t, void* ptr );
00242 
00248         static void* add_new_cut_helper(void* ptr);
00249 
00258         static int add_new_cut(
00259             float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length,
00260             uint32_t nSel, void* ptr );
00261 
00267         static void* compute_output_helper(void* ptr);
00268 
00274         static int compute_output( float64_t *output, void* ptr );
00275 
00282         static int sort( float64_t* vals, float64_t* data, uint32_t size);
00283 
00285         static inline void print(ocas_return_value_T value)
00286         {
00287               return;
00288         }
00289 
00290 
00292         inline virtual const char* get_name() const { return "WDSVMOcas"; }
00293 
00294     protected:
00303         virtual bool train_machine(CFeatures* data=NULL);
00304 
00305     protected:
00307         CStringFeatures<uint8_t>* features;
00309         bool use_bias;
00311         int32_t bufsize;
00313         float64_t C1;
00315         float64_t C2;
00317         float64_t epsilon;
00319         E_SVM_TYPE method;
00320 
00322         int32_t degree;
00324         int32_t from_degree;
00326         float32_t* wd_weights;
00328         int32_t num_vec;
00330         int32_t string_length;
00332         int32_t alphabet_size;
00333 
00335         float64_t normalization_const;
00336 
00338         float64_t bias;
00340         float64_t old_bias;
00342         int32_t* w_offsets;
00344         int32_t w_dim;
00346         int32_t w_dim_single_char;
00348         float32_t* w;
00350         float32_t* old_w;
00352         float64_t* lab;
00353 
00355         float32_t** cuts;
00357         float64_t* cp_bias;
00358 };
00359 }
00360 #endif
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines

SHOGUN Machine Learning Toolbox - Documentation