VarianceKernelNormalizer.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) 2009 Soeren Sonnenburg
00008  * Copyright (C) 2009 Fraunhofer Institute FIRST and Max-Planck-Society
00009  */
00010 
00011 #ifndef _VARIANCEKERNELNORMALIZER_H___
00012 #define _VARIANCEKERNELNORMALIZER_H___
00013 
00014 #include "kernel/KernelNormalizer.h"
00015 
00016 namespace shogun
00017 {
00027 class CVarianceKernelNormalizer : public CKernelNormalizer
00028 {
00029     public:
00032         CVarianceKernelNormalizer() : meandiff(1.0), sqrt_meandiff(1.0)
00033         {
00034         }
00035 
00037         virtual ~CVarianceKernelNormalizer()
00038         {
00039         }
00040 
00043         virtual bool init(CKernel* k)
00044         {
00045             ASSERT(k);
00046             int32_t n=k->get_num_vec_lhs();
00047             ASSERT(n>0);
00048 
00049             CFeatures* old_lhs=k->lhs;
00050             CFeatures* old_rhs=k->rhs;
00051             k->lhs=old_lhs;
00052             k->rhs=old_lhs;
00053 
00054             float64_t diag_mean=0;
00055             float64_t overall_mean=0;
00056             for (int32_t i=0; i<n; i++)
00057             {
00058                 diag_mean+=k->compute(i, i);
00059 
00060                 for (int32_t j=0; j<n; j++)
00061                     overall_mean+=k->compute(i, j);
00062             }
00063             diag_mean/=n;
00064             overall_mean/=((float64_t) n)*n;
00065 
00066             k->lhs=old_lhs;
00067             k->rhs=old_rhs;
00068 
00069             meandiff=1.0/(diag_mean-overall_mean);
00070             sqrt_meandiff=CMath::sqrt(meandiff);
00071 
00072             return true;
00073         }
00074 
00080         inline virtual float64_t normalize(
00081             float64_t value, int32_t idx_lhs, int32_t idx_rhs)
00082         {
00083             return value*meandiff;
00084         }
00085 
00090         inline virtual float64_t normalize_lhs(float64_t value, int32_t idx_lhs)
00091         {
00092             return value*sqrt_meandiff;
00093         }
00094 
00099         inline virtual float64_t normalize_rhs(float64_t value, int32_t idx_rhs)
00100         {
00101             return value*sqrt_meandiff;
00102         }
00103 
00105         inline virtual const char* get_name() const { return "VarianceKernelNormalizer"; }
00106 
00107     protected:
00109         float64_t meandiff;
00111         float64_t sqrt_meandiff;
00112 };
00113 }
00114 #endif

SHOGUN Machine Learning Toolbox - Documentation