source: trunk/test/statistics.cc @ 3236

// $Id: statistics.cc 3236 2014-05-23 13:42:51Z peter $

/*
  Copyright (C) 2004 Jari Häkkinen, Peter Johansson
  Copyright (C) 2005 Peter Johansson
  Copyright (C) 2006 Jari Häkkinen, Peter Johansson, Markus Ringnér
  Copyright (C) 2007, 2008, 2009 Jari Häkkinen, Peter Johansson
  Copyright (C) 2010, 2011, 2012, 2013 Peter Johansson

  This file is part of the yat library, http://dev.thep.lu.se/yat

  The yat library 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.

  The yat library is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with yat. If not, see <http://www.gnu.org/licenses/>.
*/

#include <config.h>

#include "Suite.h"

#include "yat/classifier/Target.h"
#include "yat/statistics/Average.h"
#include "yat/statistics/utility.h"
#include "yat/statistics/tTest.h"
#include "yat/utility/DataWeight.h"
#include "yat/utility/Matrix.h"
#include "yat/utility/MatrixWeighted.h"
#include "yat/utility/Vector.h"

#include <boost/concept_archetype.hpp>

#include <cmath>
#include <cstdlib>
#include <iostream>
#include <limits>
#include <map>
#include <vector>

using namespace theplu::yat;
void test_benjamini_hochberg(test::Suite&);
void test_benjamini_hochberg_unsorted(test::Suite&);
void test_entropy(test::Suite&);
void test_mad(test::Suite&);
void test_mutual_information(test::Suite&);

void test_median_empty(test::Suite&);
void test_percentiler(test::Suite&);
void test_percentiler_nan(test::Suite&);

template<typename RandomAccessIterator>
void test_percentiler(test::Suite&, RandomAccessIterator,
                      RandomAccessIterator,
                      double p, double correct);

template<typename RandomAccessIterator1, typename RandomAccessIterator2>
void cmp_percentiler(test::Suite&,
                     RandomAccessIterator1,
                     RandomAccessIterator1,
                     RandomAccessIterator2,
                     RandomAccessIterator2);

int main(int argc, char* argv[])
{
  test::Suite suite(argc, argv);

  utility::Vector gsl_vec(10);
  std::vector<double> data;
  for (unsigned int i=0; i<10; i++){
    data.push_back(static_cast<double>(i));
    gsl_vec(i)=i;
  }

  double m=statistics::median(data.begin(), data.end());
  double m_gsl=statistics::median(gsl_vec.begin(), gsl_vec.end());
  if (m!=4.5 || m!=m_gsl)
    suite.add(false);
  if (false) {
    using statistics::median;
    double x = median(boost::random_access_iterator_archetype<double>(),
                      boost::random_access_iterator_archetype<double>());
    test::dummie_function(x);
    x = median(boost::random_access_iterator_archetype<utility::DataWeight>(),
               boost::random_access_iterator_archetype<utility::DataWeight>());
    test::dummie_function(x);
  }
  statistics::percentile2(data.begin(), data.end(), 100);
  data.resize(1);
  statistics::median(data.begin(), data.end());
  // testing percentile2
  test_percentiler(suite);

  // test weighted percentiler with NaNs
  test_percentiler_nan(suite);

  double skewness_gsl=statistics::skewness(gsl_vec);
  if (!suite.equal(1-skewness_gsl, 1.0) )
    suite.add(false);
  double kurtosis_gsl=statistics::kurtosis(gsl_vec);
  suite.add(suite.equal_fix(kurtosis_gsl,-1.5616363636363637113,1e-10));
  statistics::Average func;
  suite.add(suite.equal(func(gsl_vec.begin(), gsl_vec.end()),4.5));
  // easiest way to get a weighted iterator
  classifier::MatrixLookupWeighted mlw(10,20,2.0, 1.0);
  suite.add(suite.equal(func(mlw.begin(), mlw.end()),2.0));
  // do not run compiler test
  if (false) {
    statistics::Average average;
    double x = average(boost::input_iterator_archetype<double>(),
                       boost::input_iterator_archetype<double>());
    test::dummie_function(x);
    using utility::DataWeight;
    x = average(boost::input_iterator_archetype_no_proxy<DataWeight>(),
                boost::input_iterator_archetype_no_proxy<DataWeight>());
    test::dummie_function(x);
  }

  test_mad(suite);

  // do not run compiler test
  if (false) {
    statistics::tTest t_test;
    classifier::Target target;
    add(t_test, boost::forward_iterator_archetype<double>(),
        boost::forward_iterator_archetype<double>(), target);
    add(t_test, boost::forward_iterator_archetype<utility::DataWeight>(),
        boost::forward_iterator_archetype<utility::DataWeight>(), target);
  }
  test_benjamini_hochberg(suite);
  test_benjamini_hochberg_unsorted(suite);
  test_entropy(suite);
  test_median_empty(suite);
  test_mutual_information(suite);
  return suite.return_value();
}


void test_benjamini_hochberg(test::Suite& suite)
{
  std::vector<double> p;
  p.push_back(0.0001);
  p.push_back(0.01);
  p.push_back(0.015);
  p.push_back(0.5);
  p.push_back(0.99);
  std::vector<double> q(p.size());
  statistics::benjamini_hochberg(p.begin(), p.end(), q.begin());
  suite.add(suite.equal(q[0], p[0]*5));
  suite.add(suite.equal(q[1], p[1]*2.5));
  suite.add(suite.equal(q[2], 0.025));
  suite.add(suite.equal(q[3], p[3]*1.25));
  suite.add(suite.equal(q[4], 0.99));

  // do nut run compiler test
  if (false) {
    using statistics::benjamini_hochberg;
    benjamini_hochberg(boost::bidirectional_iterator_archetype<double>(),
                       boost::bidirectional_iterator_archetype<double>(),
                       boost::mutable_bidirectional_iterator_archetype<double>());
  }
}


void test_benjamini_hochberg_unsorted(test::Suite& suite)
{
  std::vector<double> p;
  p.push_back(0.015);
  p.push_back(0.0001);
  p.push_back(0.01);
  p.push_back(0.5);
  p.push_back(0.99);
  std::vector<double> q(p.size());
  statistics::benjamini_hochberg_unsorted(p.begin(), p.end(), q.begin());
  suite.add(suite.equal(q[1], p[1]*5));
  suite.add(suite.equal(q[2], p[2]*2.5));
  suite.add(suite.equal(q[0], 0.025));
  suite.add(suite.equal(q[3], p[3]*1.25));
  suite.add(suite.equal(q[4], 0.99));

  // do nut run compiler test
  if (false) {
    using statistics::benjamini_hochberg_unsorted;
    boost::random_access_iterator_archetype<double> input;
    boost::mutable_random_access_iterator_archetype<double> result;
    benjamini_hochberg_unsorted(input, input, result);
  }
}


void test_entropy(test::Suite& suite)
{
  suite.out() << "testing entropy(2)\n";
  using statistics::entropy;
  std::vector<int> x(10000,0);
  x[512] = 42;
  double e = entropy(x.begin(), x.end());
  if (e>1e-15) {
    suite.add(false);
    suite.out() << "entropy: " << e << " expected close to 0\n";
  }
  x[0] = 42;
  e = entropy(x.begin(), x.end());
  if (e<=0) {
    suite.add(false);
    suite.out() << "entropy: " << e << " expected > 0\n";
  }

  // do not run compiler test
  if (false) {
    entropy(boost::input_iterator_archetype<double>(),
            boost::input_iterator_archetype<double>());
  }
}


void test_mad(test::Suite& suite)
{
  suite.err() << "testing mad" << std::endl;
  utility::Vector x(3);
  x(0) = 3;
  x(1) = 1;
  x(2) = 100;
  suite.add(suite.equal(statistics::mad(x.begin(), x.end()), 2));

  std::vector<utility::DataWeight> wx(3);
  wx[0] = utility::DataWeight(3, 0.4);
  wx[1] = utility::DataWeight(1, 0.4);
  wx[2] = utility::DataWeight(100, 0.6);
  suite.add(suite.equal(statistics::mad(wx.begin(), wx.end()), 2));
  // do not run compiler test
  if (false) {
    using statistics::mad;
    double x = mad(boost::random_access_iterator_archetype<double>(),
                   boost::random_access_iterator_archetype<double>());
    test::dummie_function(x);
    x = mad(boost::random_access_iterator_archetype<utility::DataWeight>(),
            boost::random_access_iterator_archetype<utility::DataWeight>());
    test::dummie_function(x);
  }
}


void test_mutual_information(test::Suite& suite)
{
  suite.out() << "testing mutual_information\n";
  using statistics::mutual_information;
  utility::Matrix x(2,2);
  x(0,0) = 100;
  x(1,1) = 100;
  double mi = mutual_information(x);
  if (!suite.add(suite.equal(mi,1.0,100))) {
    suite.err() << "error: mutual information: " << mi << "\n";
  }

  // testing a non-square Matrix
  x.resize(3,4,0);
  x(0,0) = 1;
  x(1,1) = 1;
  x(2,2) = 1;
  x(2,3) = 1;
  mi = mutual_information(x);
  suite.out() << "mi: " << mi << "\n";
}


// test for ticket #660
void test_median_empty(test::Suite& suite)
{
  std::vector<double> x;
  double m = 0;
  m = statistics::median(x.begin(), x.end(), true);
  test::dummie_function(m);
}


void test_percentiler(test::Suite& suite)
{
  suite.err() << "testing unweighted percentile2" << std::endl;
  std::vector<double> x;
  x.reserve(6);
  for (unsigned int i=0; i<5; i++){
    x.push_back(static_cast<double>(i+1));
  }
  test_percentiler(suite, x.begin(), x.end(), 50, 3);
  x.push_back(6);
  test_percentiler(suite, x.begin(), x.end(), 50, 3.5);
  test_percentiler(suite, x.begin(), x.end(), 25, 2);
  test_percentiler(suite, x.begin(), x.end(), 0, 1);
  test_percentiler(suite, x.begin(), x.end(), 10, 1);

  suite.err() << "testing duplication of data\n";
  std::vector<double> x2(x);
  for (size_t i=0; i<x.size(); ++i)
    x2.push_back(x[i]);
  cmp_percentiler(suite, x.begin(), x.end(), x2.begin(), x2.end());


  // testing weighted

  suite.err() << "testing weighted percentile2" << std::endl;
  std::vector<utility::DataWeight> xw(x.size());
  for (size_t i=0; i<xw.size(); ++i) {
    xw[i].data() = x[i];
    xw[i].weight() = 1.0;
  }
  const std::vector<utility::DataWeight> xw_orig(xw);
  suite.err() << "testing weighted" << std::endl;
  test_percentiler(suite, xw.begin(), xw.end(), 0, 1);
  test_percentiler(suite, xw.begin(), xw.end(), 100, 6);
  test_percentiler(suite, xw.begin(), xw.end(), 49, 3);
  test_percentiler(suite, xw.begin(), xw.end(), 51, 4);
  test_percentiler(suite, xw.begin(), xw.end(), 50, 3.5);
  test_percentiler(suite, x.begin(), x.end(), 10, 1);

  suite.err() << "testing weighted with unity weights" << std::endl;
  cmp_percentiler(suite, x.begin(), x.end(), xw.begin(), xw.end());

  suite.err() << "testing that w=0 equals removed data point\n";
  xw=xw_orig;
  std::vector<utility::DataWeight> xw2(xw_orig);
  xw[3].weight() = 0.0;
  xw2.erase(xw2.begin()+3);
  cmp_percentiler(suite, xw.begin(), xw.end(), xw2.begin(), xw2.end());

  suite.err() << "testing rescaling of weights\n";
  xw2 = xw;
  for (size_t i=0; i<xw2.size(); ++i)
    xw2[i].weight()*=2;
  cmp_percentiler(suite, xw.begin(), xw.end(), xw2.begin(), xw2.end());

  // do not run compiler test
  if (false) {
    statistics::Percentiler percentiler(50);
    using boost::random_access_iterator_archetype;
    typedef random_access_iterator_archetype<double> Iterator;
    double x = percentiler(Iterator(), Iterator()); 
    test::dummie_function(x);
    using utility::DataWeight;
    typedef random_access_iterator_archetype<DataWeight> WeigtedItererator;
    x = percentiler(WeigtedItererator(), WeigtedItererator());
    test::dummie_function(x);
  }
}

void test_percentiler_nan(test::Suite& suite)
{
  using utility::DataWeight;
  std::vector<double> v;
  v.push_back(1);
  v.push_back(10);
  v.push_back(4);
  v.push_back(2);
  std::vector<DataWeight> wv(5);
  wv[0] = DataWeight(v[0]);
  wv[1] = DataWeight(v[1]);
  wv[2] = DataWeight(std::numeric_limits<double>::quiet_NaN(), 0.0);
  wv[3] = DataWeight(v[2]);
  wv[4] = DataWeight(v[3]);

  cmp_percentiler(suite, v.begin(), v.end(), wv.begin(), wv.end());
}

template<typename RandomAccessIterator>
void test_percentiler(test::Suite& suite, 
                      RandomAccessIterator first, 
                      RandomAccessIterator last,
                      double p, double correct)
{
  using statistics::percentile2;
  double x = percentile2(first, last, p);
  if (!suite.add(suite.equal(x, correct, 10))) {
    suite.err() << "Error in percentile2 for " << p << "th percentile \n";
    suite.err() << "  calculated value: " << x << "\n";
    suite.err() << "  expected value: " << correct << "\n";
  }
}

template<typename RandomAccessIterator1, typename RandomAccessIterator2>
void cmp_percentiler(test::Suite& suite, 
                     RandomAccessIterator1 first1, 
                     RandomAccessIterator1 last1,
                     RandomAccessIterator2 first2,
                     RandomAccessIterator2 last2)
{
  for (double p=0; p<=100; p+=10) {
    double correct=statistics::percentile2(first1, last1, p);
    test_percentiler(suite, first2, last2, p, correct);
  }

}