contrib/mul/mbl/mbl_sample_stats_1d.cxx

Go to the documentation of this file.

#include "mbl_sample_stats_1d.h"
//:
// \file
#include <vsl/vsl_vector_io.h>
#include <vcl_cassert.h>
#include <vcl_cmath.h>
#include <vcl_limits.h>
#include <vcl_algorithm.h>


//=========================================================================
mbl_sample_stats_1d::mbl_sample_stats_1d(const vcl_vector<double> &samples)
{
  clear();
  for (unsigned i=0, n=samples.size(); i<n; ++i)
  {
    add_sample(samples[i]);
  }
}


//=========================================================================
mbl_sample_stats_1d::mbl_sample_stats_1d(const vnl_vector<double> &samples)
{
  clear();
  for (unsigned i=0, n=samples.size(); i<n; ++i)
  {
    add_sample(samples[i]);
  }
}


//=========================================================================
mbl_sample_stats_1d::mbl_sample_stats_1d()
{
  clear();
}


//=========================================================================
mbl_sample_stats_1d::~mbl_sample_stats_1d()
{
}


//=========================================================================
void mbl_sample_stats_1d::clear()
{
  samples_.resize(0);
  stats_1d_.clear();
  use_mvue_=true;
}


//=========================================================================
void mbl_sample_stats_1d::add_sample(double v)
{
  stats_1d_.obs(v);
  samples_.push_back(v);
  return;
}


//=========================================================================
unsigned mbl_sample_stats_1d::n_samples() const
{
  return samples_.size();
}


//=========================================================================
double mbl_sample_stats_1d::mean() const
{
  return stats_1d_.mean();
}


//=========================================================================
double mbl_sample_stats_1d::mean_of_absolutes() const
{
  double abs_sum = 0;
  for (unsigned i=0, n=samples_.size(); i<n; ++i)
    abs_sum+=vcl_fabs(samples_[i]);
  return abs_sum/samples_.size();
}


//=========================================================================
double mbl_sample_stats_1d::median() const
{
  //     return nth_percentile(50);
  double ret;

  if (samples_.size()>0)
  {
    if ( samples_.size() % 2 == 0 )
    {
      unsigned index = samples_.size() / 2 - 1;

      vcl_vector<double> tmp=samples_;
      vcl_vector<double>::iterator index_it0 = tmp.begin() + index;
      vcl_nth_element(tmp.begin(),index_it0,tmp.end(),vcl_less<double>());
      vcl_vector<double>::iterator index_it1 = tmp.begin() + index + 1;
      vcl_nth_element(tmp.begin(),index_it1,tmp.end(),vcl_less<double>());

      ret = (*index_it0 + *index_it1);
      ret /= 2.0;
    }
    else
    {
      unsigned index = (samples_.size() - 1) / 2;

      vcl_vector<double> tmp=samples_;
      vcl_vector<double>::iterator index_it = tmp.begin() + index;
      vcl_nth_element(tmp.begin(),index_it,tmp.end(),vcl_less<double>());

      ret = *index_it;
    }
  }
  else // crazy value if  no samples
  {
    ret = vcl_numeric_limits<double>::max();
  }
  return ret;
}


//=========================================================================
double mbl_sample_stats_1d::quantile(double q) const
{
  const unsigned n = samples_.size();

  // These checks are only asserts because client code is responsible for avoiding these errors.
  assert(q>=0.0 && q<=1.0);
  assert(n>0);

  // Map the specified quantile to a real-valued "index", i.e. a float lying between 2 integer indices
  double float_index = (n-1)*q;

  // Get the integer index immediately below (and enforce the bounds)
  double f0 = vcl_floor(float_index);
  f0 = f0<0.0 ? 0.0 : f0>n-1.0 ? n-1.0 : f0;
  unsigned i0 = static_cast<unsigned>(f0);

  // Get the integer index immediately above (and enforce the bounds)
  double f1 = vcl_ceil(float_index);
  f1 = f1<0.0 ? 0.0 : f1>n-1.0 ? n-1.0 : f1;
  unsigned i1 = static_cast<unsigned>(f1);

  // Get the 2 values bracketing the specified quantile position
  vcl_vector<double> tmp = samples_;
  vcl_vector<double>::iterator index_it0 = tmp.begin() + i0;
  vcl_nth_element(tmp.begin(), index_it0, tmp.end(), vcl_less<double>());
  vcl_vector<double>::iterator index_it1 = tmp.begin() + i1;
  vcl_nth_element(tmp.begin(), index_it1, tmp.end(), vcl_less<double>());

  // Linearly interpolate between the 2 values
  double f = float_index - f0;
  double ret = (1.0-f)*(*index_it0) + (f)*(*index_it1);
  return ret;
}


//=========================================================================
double mbl_sample_stats_1d::nth_percentile(int n) const
{
  if (samples_.size()==0)
    return vcl_numeric_limits<double>::max();

  double fact = double(n)/100.0;
  int index=int(fact*(samples_.size()-1));
  vcl_vector<double> tmp=samples_;
  vcl_vector<double>::iterator index_it = tmp.begin() + index;
  vcl_nth_element(tmp.begin(),index_it,tmp.end(),vcl_less<double>());
  double ret = *index_it;
  return ret;
}


//=========================================================================
double mbl_sample_stats_1d::variance() const
{
  double v=0;

  if (samples_.size()>1)
  {
    double mean_v = mean();
    double sum_sq = sum_squares();
    v = sum_sq - samples_.size()*(mean_v * mean_v);

    if (use_mvue_)
    {
      v /= (samples_.size()-1);
    }
    else
    {
      v /= samples_.size();
    }
  }

  return v;
}


//=========================================================================
double mbl_sample_stats_1d::sd() const
{
  return vcl_sqrt(variance());
}


//=========================================================================
double mbl_sample_stats_1d::stdError() const
{
  double se = variance();
  if (use_mvue_)
  {
    se /= samples_.size()-1;
  }
  else
  {
    se /= samples_.size();
  }

  return vcl_sqrt(se);
}


//=========================================================================
double mbl_sample_stats_1d::skewness() const
{
  double skew = 0;

  // skew
  // calculated as
  // ( Sum_i (Y_i-MEAN)^3 ) / ((N-1)*sigma^3)
  // where N is the number of samples
  // sigma is the standard deviation

  if (samples_.size()>1)
  {
    double s=sd();
    double m=mean();

    for (unsigned i=0, n=samples_.size(); i<n; ++i)
    {
      double tmp=samples_[i]-m;
      skew += (tmp*tmp*tmp) ;
    }

    skew /= ( (samples_.size()-1) * s * s * s );
  }

  return skew;
}


//=========================================================================
double mbl_sample_stats_1d::kurtosis() const
{
  double kurt = 0;

  // kurtosis
  // calculated as
  // -3 + ( Sum_i (Y_i-MEAN)^4 ) / ((N-1)*sigma^4)
  // where N is the number of samples
  // sigma is the standard deviation

  if (samples_.size()>1)
  {
    double s=sd();
    double m=mean();

    for (unsigned i=0, n=samples_.size(); i<n; ++i)
    {
      double tmp=samples_[i]-m;
      kurt += (tmp*tmp*tmp*tmp) ;
    }

    kurt /= ( (samples_.size()-1) * s * s * s *s);
    kurt -= 3;
  }
  return kurt;
}


//=========================================================================
double mbl_sample_stats_1d::min() const
{
  if (samples_.size()==0) return vcl_numeric_limits<double>::max();
  else return stats_1d_.min();
}


//=========================================================================
double mbl_sample_stats_1d::max() const
{
  if (samples_.size()==0) return vcl_numeric_limits<double>::min();
  else return stats_1d_.max();
}


//=========================================================================
double mbl_sample_stats_1d::sum() const
{
  return stats_1d_.sum();
}


//=========================================================================
double mbl_sample_stats_1d::sum_squares() const
{
  return stats_1d_.sumSq();
}


//=========================================================================
double mbl_sample_stats_1d::rms() const
{
  double ms=sum_squares()/stats_1d_.nObs();
  return vcl_sqrt( ms );
}


//=========================================================================
mbl_sample_stats_1d& mbl_sample_stats_1d::operator+=(const mbl_sample_stats_1d& s1)
{
  // add new samples
  for (unsigned i=0;i<s1.samples().size();++i)
  {
    add_sample(s1.samples()[i]);
  }

  return *this ;
}


//=========================================================================
// Test for equality
bool mbl_sample_stats_1d::operator==(const mbl_sample_stats_1d& s) const
{
  return samples_==s.samples_ && use_mvue_==s.use_mvue_;
}

// =============================================
short mbl_sample_stats_1d::version_no() const
{
  return 1;
}


//=========================================================================
void mbl_sample_stats_1d::b_write(vsl_b_ostream& bfs) const
{
  vsl_b_write(bfs,version_no());
  vsl_b_write(bfs,samples_);
  vsl_b_write(bfs,stats_1d_);
  vsl_b_write(bfs,use_mvue_);
}


//=========================================================================
void mbl_sample_stats_1d::b_read(vsl_b_istream& bfs)
{
  if (!bfs) return;

  short file_version_no;
  vsl_b_read(bfs,file_version_no);

  switch (file_version_no)
  {
  case 1:
    vsl_b_read(bfs,samples_);
    vsl_b_read(bfs,stats_1d_);
    vsl_b_read(bfs,use_mvue_);
    break;
  default :
    vcl_cerr << "I/O ERROR: mbl_sample_stats_1d::b_read(vsl_b_istream&)\n"
             << "           Unknown version number "<< file_version_no << '\n';
    bfs.is().clear(vcl_ios::badbit); // Set an unrecoverable IO error on stream
    return;
  }
}


//=========================================================================
void mbl_sample_stats_1d::print_summary(vcl_ostream& os) const
{
  os << "mbl_sample_stats_1d: ";
  if (samples_.size()==0)
  {
    os << "No samples.";
  }
  else
  {
    os << "mean: "<< mean()
       << " use MVUE: "<< use_mvue_
       << " sd: "<< sd()
       << " ["<<stats_1d_.min()<<','<<stats_1d_.max()<<"] N:"<<samples_.size();
  }
}


//=========================================================================
// Print all data
void mbl_sample_stats_1d::print_all(vcl_ostream& os,
                                    const vcl_string& delim/*="\n"*/) const
{
  unsigned nSamples = samples_.size();
  for (unsigned i=0; i<nSamples; ++i)
  {
    os << samples_[i] << delim;
  }
}


//=========================================================================
vcl_ostream& operator<<(vcl_ostream& os, const mbl_sample_stats_1d& stats)
{
  stats.print_summary(os);
  return os;
}


//=========================================================================
// Stream output operator for class reference
void vsl_print_summary(vcl_ostream& os,const mbl_sample_stats_1d& stats)
{
  stats.print_summary(os);
}


//=========================================================================
// Print all data
void vsl_print_all(vcl_ostream& os, const mbl_sample_stats_1d& stats)
{
  stats.print_all(os);
}


//=========================================================================
mbl_sample_stats_1d operator+(const mbl_sample_stats_1d& s1, const mbl_sample_stats_1d& s2)
{
  mbl_sample_stats_1d r = s1;
  r+=s2;

  return r;
}


//=========================================================================
// Binary file stream output operator for class reference
void vsl_b_write(vsl_b_ostream& bfs, const mbl_sample_stats_1d& b)
{
  b.b_write(bfs);
}


//=========================================================================
// Binary file stream input operator for class reference
void vsl_b_read(vsl_b_istream& bfs, mbl_sample_stats_1d& b)
{
  b.b_read(bfs);
}