Labels.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) 1999-2009 Soeren Sonnenburg
 * Written (W) 1999-2008 Gunnar Raetsch
 * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#include "features/Labels.h"
#include "lib/common.h"
#include "lib/File.h"
#include "lib/io.h"
#include "lib/Mathematics.h"

#ifdef HAVE_BOOST_SERIALIZATION
#include <boost/serialization/export.hpp>
BOOST_CLASS_EXPORT(shogun::CLabels);
#endif //HAVE_BOOST_SERIALIZATION

using namespace shogun;

CLabels::CLabels()
: CSGObject()
{
    labels = NULL;
    num_labels = 0;
    m_confidences=NULL;
    m_num_classes=0;
}

CLabels::CLabels(int32_t num_lab)
: CSGObject(), num_labels(num_lab)
{
    labels=new float64_t[num_lab];
    for (int32_t i=0; i<num_lab; i++)
        labels[i]=0;

    m_num_classes=0;
    m_confidences=NULL;
}

CLabels::CLabels(float64_t* p_labels, int32_t len)
: CSGObject()
{
    labels = NULL;
    num_labels = 0;

    set_labels(p_labels, len);
    
    // We don't allocate the confidences matrix, unless it is necessary. 
    // For problems with many classes and samples it might get really big.
    m_num_classes=get_num_classes();
    m_confidences=NULL; 
}

CLabels::CLabels(float64_t* in_confidences, int32_t in_num_labels, 
                 int32_t in_num_classes)
: CSGObject()
{
    labels=new float64_t[in_num_labels];
    for (int32_t i=0; i<in_num_labels; i++)
        labels[i]=0;

    m_num_classes=in_num_classes;
    m_confidences=in_confidences;
    find_labels();
}

CLabels::CLabels(CFile* loader)
: CSGObject()
{
    num_labels=0;
    labels=NULL;
    m_num_classes=0;
    m_confidences=NULL;

    load(loader);
}

CLabels::~CLabels()
{
    delete[] labels;
    delete[] m_confidences;
    num_labels=0;
    m_num_classes=0;
    labels=NULL;
    m_confidences=NULL;
}

void CLabels::set_labels(float64_t* p_labels, int32_t len)
{
    ASSERT(len>0);
    num_labels=len;

    delete[] labels;
    labels=CMath::clone_vector(p_labels, len);
}

void CLabels::set_confidences(float64_t* in_confidences, int32_t in_num_labels, 
                              int32_t in_num_classes)
{
    if (num_labels && (num_labels != in_num_labels))
    {
        SG_ERROR("Shape of confidence matrix mismatch (number of "
                "labels = %d does not match %d\n", num_labels, in_num_labels);
    }

    if (m_num_classes && (m_num_classes != in_num_classes))
    {
        SG_ERROR("Shape of confidence matrix mismatch (number of "
                "num_classes = %d does not match %d\n", m_num_classes, in_num_classes);
    }

    delete[] m_confidences;

    num_labels=in_num_labels;
    m_num_classes=in_num_classes;
    m_confidences=in_confidences;
    find_labels();
}

float64_t* CLabels::get_confidences(int32_t& out_num_labels, int32_t& out_num_classes)
{
    out_num_labels=num_labels;
    out_num_classes=m_num_classes;
    
    if (!num_labels || !m_num_classes || !m_confidences)
        SG_ERROR("No labels / confidences set\n");

    float64_t* out_conf=new float64_t[num_labels*m_num_classes];
    memcpy(out_conf, m_confidences, num_labels*m_num_classes*sizeof(float64_t));
    return out_conf;
}

void CLabels::get_confidences(float64_t** dst, int32_t* out_num_labels, int32_t* out_num_classes)
{
    ASSERT(dst && out_num_labels && out_num_classes);

    if (num_labels<=0 || m_num_classes<=0 || !m_confidences)
        SG_ERROR("No labels / confidences set\n");

    *dst=NULL;
    *out_num_labels=num_labels;
    *out_num_classes=m_num_classes;

    float64_t* out_conf= (float64_t*) malloc((size_t) sizeof(float64_t)*num_labels*m_num_classes);
    memcpy(out_conf, m_confidences, num_labels*m_num_classes*sizeof(float64_t));
    *dst=out_conf;
}

float64_t* CLabels::get_sample_confidences(const int32_t& in_sample_index, 
                                           int32_t& out_num_classes)
{
    out_num_classes=m_num_classes;

    if (!(in_sample_index>=0 && in_sample_index<num_labels &&
                m_num_classes && m_confidences))
    {
        SG_ERROR("No labels / confidences set\n");
    }

    float64_t* out_conf=new float64_t[m_num_classes];
    for (int32_t n_class=0; n_class<m_num_classes; n_class++)
    {
        out_conf[n_class]=m_confidences[n_class+in_sample_index*m_num_classes];
    }
    return out_conf;
}

void CLabels::find_labels()
{
    ASSERT(m_confidences);
    ASSERT(labels);
    
    float64_t max_conf;
    int32_t index;
    for (int32_t n_samp=0; n_samp<num_labels; n_samp++)
    {
        max_conf=m_confidences[n_samp];
        labels[n_samp]=0;
        for (int32_t n_class=1; n_class<m_num_classes; n_class++)
        {
            index=n_samp+n_class*m_num_classes;
            if (m_confidences[index]>max_conf)
            {
                max_conf=m_confidences[index];
                labels[n_samp]=n_class;             
            }
        }
    }
}

bool CLabels::is_two_class_labeling()
{
    ASSERT(labels);
    bool found_plus_one=false;
    bool found_minus_one=false;

    for (int32_t i=0; i<num_labels; i++)
    {
        if (labels[i]==+1.0)
            found_plus_one=true;
        else if (labels[i]==-1.0)
            found_minus_one=true;
        else
            SG_ERROR("Not a two class labeling label[%d]=%f (only +1/-1 allowed)\n", i, labels[i]);
    }

    if (!found_plus_one)
        SG_ERROR("Not a two class labeling - no positively labeled examples found\n");
    if (!found_minus_one)
        SG_ERROR("Not a two class labeling - no negatively labeled examples found\n");

    return true;
}

int32_t CLabels::get_num_classes()
{
    int32_t n=-1;
    int32_t* lab=get_int_labels(n);

    int32_t num_classes=0;
    for (int32_t i=0; i<n; i++)
        num_classes=CMath::max(num_classes,lab[i]);

    delete[] lab;

    return num_classes+1;
}

float64_t* CLabels::get_labels(int32_t &len)
{
    len=num_labels;

    if (num_labels>0)
    {
        float64_t* _labels=new float64_t[num_labels] ;
        for (int32_t i=0; i<len; i++)
            _labels[i]=get_label(i) ;
        return _labels ;
    }
    else 
        return NULL;
}

void CLabels::get_labels(float64_t** p_labels, int32_t* len)
{
    ASSERT(p_labels && len);
    *p_labels=NULL;
    *len=num_labels;

    if (num_labels>0)
    {
        *p_labels=(float64_t*) malloc(sizeof(float64_t)*num_labels);

        for (int32_t i=0; i<num_labels; i++)
            (*p_labels)[i]=get_label(i);
    }
}

int32_t* CLabels::get_int_labels(int32_t &len)
{
    len=num_labels;

    if (num_labels>0)
    {
        int32_t* _labels=new int32_t[num_labels] ;
        for (int32_t i=0; i<len; i++)
            _labels[i]= (int32_t) get_label(i) ;
        return _labels ;
    }
    else 
        return NULL;
}

void CLabels::set_int_labels(int32_t * mylabels, int32_t len)
{
    num_labels = len ;
    delete[] labels ;
    
    labels = new float64_t[num_labels] ;
    for (int32_t i=0; i<num_labels; i++)
        set_int_label(i, mylabels[i]) ;
}

void CLabels::load(CFile* loader)
{
    delete[] labels;
    delete[] m_confidences;
    m_confidences = NULL;
    num_labels=0;
    ASSERT(loader);
    loader->get_real_vector(labels, num_labels);
    m_num_classes=get_num_classes();
}

void CLabels::save(CFile* writer)
{
    ASSERT(writer);
    ASSERT(labels && labels>0);
    writer->set_real_vector(labels, num_labels);
}