PruneVarSubMean.cpp

Go to the documentation of this file.

/*
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * Written (W) 1999-2008 Gunnar Raetsch
 * Written (W) 1999-2009 Soeren Sonnenburg
 * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#include <shogun/preprocessor/PruneVarSubMean.h>
#include <shogun/preprocessor/DensePreprocessor.h>
#include <shogun/features/Features.h>
#include <shogun/io/SGIO.h>
#include <shogun/mathematics/Math.h>

using namespace shogun;

CPruneVarSubMean::CPruneVarSubMean(bool divide)
: CDensePreprocessor<float64_t>(), idx(NULL), mean(NULL),
    std(NULL), num_idx(0), divide_by_std(divide), initialized(false)
{
}

CPruneVarSubMean::~CPruneVarSubMean()
{
    cleanup();
}

bool CPruneVarSubMean::init(CFeatures* features)
{
    if (!initialized)
    {
        ASSERT(features->get_feature_class()==C_DENSE);
        ASSERT(features->get_feature_type()==F_DREAL);

        CDenseFeatures<float64_t>* simple_features=(CDenseFeatures<float64_t>*) features;
        int32_t num_examples = simple_features->get_num_vectors();
        int32_t num_features = simple_features->get_num_features();

        SG_FREE(mean);
        SG_FREE(idx);
        SG_FREE(std);
        mean=NULL;
        idx=NULL;
        std=NULL;

        mean=SG_MALLOC(float64_t, num_features);
        float64_t* var=SG_MALLOC(float64_t, num_features);
        int32_t i,j;

        for (i=0; i<num_features; i++)
        {
            mean[i]=0;
            var[i]=0 ;
        }

        SGMatrix<float64_t> feature_matrix = simple_features->get_feature_matrix();

        // compute mean
        for (i=0; i<num_examples; i++)
        {
            for (j=0; j<num_features; j++)
                mean[j]+=feature_matrix.matrix[i*num_features+j];
        }

        for (j=0; j<num_features; j++)
            mean[j]/=num_examples;

        // compute var
        for (i=0; i<num_examples; i++)
        {
            for (j=0; j<num_features; j++)
                var[j]+=CMath::sq(mean[j]-feature_matrix.matrix[i*num_features+j]);
        }

        int32_t num_ok=0;
        int32_t* idx_ok=SG_MALLOC(int, num_features);

        for (j=0; j<num_features; j++)
        {
            var[j]/=num_examples;

            if (var[j]>=1e-14)
            {
                idx_ok[num_ok]=j;
                num_ok++ ;
            }
        }

        SG_INFO( "Reducing number of features from %i to %i\n", num_features, num_ok) ;

        SG_FREE(idx);
        idx=SG_MALLOC(int, num_ok);
        float64_t* new_mean=SG_MALLOC(float64_t, num_ok);
        std=SG_MALLOC(float64_t, num_ok);

        for (j=0; j<num_ok; j++)
        {
            idx[j]=idx_ok[j] ;
            new_mean[j]=mean[idx_ok[j]];
            std[j]=sqrt(var[idx_ok[j]]);
        }
        num_idx = num_ok ;
        SG_FREE(idx_ok);
        SG_FREE(mean);
        SG_FREE(var);
        mean = new_mean;

        initialized = true;
        return true;
    }
    else
        return false;
}

void CPruneVarSubMean::cleanup()
{
    SG_FREE(idx);
    idx=NULL;
    SG_FREE(mean);
    mean=NULL;
    SG_FREE(std);
    std=NULL;
}

SGMatrix<float64_t> CPruneVarSubMean::apply_to_feature_matrix(CFeatures* features)
{
    ASSERT(initialized);

    int32_t num_vectors=0;
    int32_t num_features=0;
    float64_t* m=((CDenseFeatures<float64_t>*) features)->get_feature_matrix(num_features, num_vectors);

    SG_INFO( "get Feature matrix: %ix%i\n", num_vectors, num_features);
    SG_INFO( "Preprocessing feature matrix\n");
    for (int32_t vec=0; vec<num_vectors; vec++)
    {
        float64_t* v_src=&m[num_features*vec];
        float64_t* v_dst=&m[num_idx*vec];

        if (divide_by_std)
        {
            for (int32_t feat=0; feat<num_idx; feat++)
                v_dst[feat]=(v_src[idx[feat]]-mean[feat])/std[feat];
        }
        else
        {
            for (int32_t feat=0; feat<num_idx; feat++)
                v_dst[feat]=(v_src[idx[feat]]-mean[feat]);
        }
    }

    ((CDenseFeatures<float64_t>*) features)->set_num_features(num_idx);
    ((CDenseFeatures<float64_t>*) features)->get_feature_matrix(num_features, num_vectors);
    SG_INFO( "new Feature matrix: %ix%i\n", num_vectors, num_features);

    return ((CDenseFeatures<float64_t>*) features)->get_feature_matrix();
}

SGVector<float64_t> CPruneVarSubMean::apply_to_feature_vector(SGVector<float64_t> vector)
{
    float64_t* ret=NULL;

    if (initialized)
    {
        ret=SG_MALLOC(float64_t, num_idx);

        if (divide_by_std)
        {
            for (int32_t i=0; i<num_idx; i++)
                ret[i]=(vector.vector[idx[i]]-mean[i])/std[i];
        }
        else
        {
            for (int32_t i=0; i<num_idx; i++)
                ret[i]=(vector.vector[idx[i]]-mean[i]);
        }
    }
    else
    {
        ret=SG_MALLOC(float64_t, vector.vlen);
        for (int32_t i=0; i<vector.vlen; i++)
            ret[i]=vector.vector[i];
    }

    return SGVector<float64_t>(ret,num_idx);
}