nojhan/paradiseo
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

update metric

Browse source 
git-svn-id: svn://scm.gforge.inria.fr/svnroot/paradiseo@264 331e1502-861f-0410-8da2-ba01fb791d7f
This commit is contained in:
liefooga 2007-04-17 15:47:45 +00:00
commit e926d39359
5 changed files with 570 additions and 358 deletions
Download patch file Download diff file Expand all files Collapse all files

60
trunk/paradiseo-moeo/src/metric/moeoBinaryMetricSavingUpdater.h
View file

@ -2,7 +2,7 @@
//----------------------------------------------------------------------------- //-----------------------------------------------------------------------------
// moeoBinaryMetricSavingUpdater.h // moeoBinaryMetricSavingUpdater.h
// (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2006 // (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2007
/* /*
This library... This library...
@ -20,17 +20,18 @@
#include <metric/moeoMetric.h> #include <metric/moeoMetric.h>
/** /**
* This class allows to save the progression of a binary metric comparing the fitness values of the current population (or archive) * This class allows to save the progression of a binary metric comparing the objective vectors of the current population (or archive)
* with the fitness values of the population (or archive) of the generation (n-1) into a file * with the objective vectors of the population (or archive) of the generation (n-1) into a file
*/ */
template < class EOT > class moeoBinaryMetricSavingUpdater:public eoUpdater template < class MOEOT >
class moeoBinaryMetricSavingUpdater : public eoUpdater
{ {
public: public:
/** /**
* The fitness type of a solution * The objective vector type of a solution
*/ */
typedef typename EOT::Fitness EOFitness; typedef typename MOEOT::ObjectiveVector ObjectiveVector;
/** /**
* Ctor * Ctor
@ -38,37 +39,30 @@ public:
* @param _pop the main population * @param _pop the main population
* @param _filename the target filename * @param _filename the target filename
*/ */
moeoBinaryMetricSavingUpdater (moeoVectorVsVectorBM < EOT, double >&_metric, moeoBinaryMetricSavingUpdater (moeoVectorVsVectorBinaryMetric < ObjectiveVector, double > & _metric, const eoPop < MOEOT > & _pop, std::string _filename) :
const eoPop < EOT > &_pop, metric(_metric), pop(_pop), filename(_filename), counter(1)
std::string _filename):metric (_metric), {}
pop (_pop), filename (_filename), counter (1)
{
}
/** /**
* Saves the metric's value for the current generation * Saves the metric's value for the current generation
*/ */
void operator () () void operator()() {
{ if (pop.size()) {
if (pop.size ()) if (firstGen) {
{
if (firstGen)
{
firstGen = false; firstGen = false;
} }
else else {
{
// creation of the two Pareto sets // creation of the two Pareto sets
std::vector < EOFitness > from; std::vector < ObjectiveVector > from;
std::vector < EOFitness > to; std::vector < ObjectiveVector > to;
for (unsigned i = 0; i < pop.size (); i++) for (unsigned i=0; i<pop.size(); i++)
from.push_back (pop[i].fitness ()); from.push_back(pop[i].objectiveVector());
for (unsigned i = 0; i < oldPop.size (); i++) for (unsigned i=0 ; i<oldPop.size(); i++)
to.push_back (oldPop[i].fitness ()); to.push_back(oldPop[i].objectiveVector());
// writing the result into the file // writing the result into the file
std::ofstream f (filename.c_str (), std::ios::app); std::ofstream f (filename.c_str(), std::ios::app);
f << counter++ << ' ' << metric (from, to) << std::endl; f << counter++ << ' ' << metric(from,to) << std::endl;
f.close (); f.close();
} }
oldPop = pop; oldPop = pop;
} }
@ -77,11 +71,11 @@ public:
private: private:
/** binary metric comparing two Pareto sets */ /** binary metric comparing two Pareto sets */
moeoVectorVsVectorBM < EOT, double >&metric; moeoVectorVsVectorBinaryMetric < ObjectiveVector, double > & metric;
/** main population */ /** main population */
const eoPop < EOT > &pop; const eoPop < MOEOT > & pop;
/** (n-1) population */ /** (n-1) population */
eoPop < EOT > oldPop; eoPop< MOEOT > oldPop;
/** target filename */ /** target filename */
std::string filename; std::string filename;
/** is it the first generation ? */ /** is it the first generation ? */
@ -91,4 +85,4 @@ private:
}; };
#endif /*MOEOBINARYMETRICSAVINGUPDATER_H_ */ #endif /*MOEOBINARYMETRICSAVINGUPDATER_H_*/

70
trunk/paradiseo-moeo/src/metric/moeoContributionMetric.h
View file

@ -2,7 +2,7 @@
//----------------------------------------------------------------------------- //-----------------------------------------------------------------------------
// moeoContributionMetric.h // moeoContributionMetric.h
// (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2006 // (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2007
/* /*
This library... This library...
@ -17,32 +17,24 @@
/** /**
* The contribution metric evaluates the proportion of non-dominated solutions given by a Pareto set relatively to another Pareto set * The contribution metric evaluates the proportion of non-dominated solutions given by a Pareto set relatively to another Pareto set
*
* (Meunier, Talbi, Reininger: 'A multiobjective genetic algorithm for radio network optimization', in Proc. of the 2000 Congress on Evolutionary Computation, IEEE Press, pp. 317-324) * (Meunier, Talbi, Reininger: 'A multiobjective genetic algorithm for radio network optimization', in Proc. of the 2000 Congress on Evolutionary Computation, IEEE Press, pp. 317-324)
*/ */
template < class EOT > class moeoContributionMetric:public moeoVectorVsVectorBM < EOT, template < class ObjectiveVector >
double > class moeoContributionMetric : public moeoVectorVsVectorBinaryMetric < ObjectiveVector, double >
{ {
public: public:
/**
* The fitness type of a solution
*/
typedef typename EOT::Fitness EOFitness;
/** /**
* Returns the contribution of the Pareto set '_set1' relatively to the Pareto set '_set2' * Returns the contribution of the Pareto set '_set1' relatively to the Pareto set '_set2'
* @param _set1 the first Pareto set * @param _set1 the first Pareto set
* @param _set2 the second Pareto set * @param _set2 the second Pareto set
*/ */
double operator () (const std::vector < EOFitness > &_set1, double operator()(const std::vector < ObjectiveVector > & _set1, const std::vector < ObjectiveVector > & _set2) {
const std::vector < EOFitness > &_set2) unsigned c = card_C(_set1, _set2);
{ unsigned w1 = card_W(_set1, _set2);
unsigned c = card_C (_set1, _set2); unsigned n1 = card_N(_set1, _set2);
unsigned w1 = card_W (_set1, _set2); unsigned w2 = card_W(_set2, _set1);
unsigned n1 = card_N (_set1, _set2); unsigned n2 = card_N(_set2, _set1);
unsigned w2 = card_W (_set2, _set1);
unsigned n2 = card_N (_set2, _set1);
return (double) (c / 2.0 + w1 + n1) / (c + w1 + n1 + w2 + n2); return (double) (c / 2.0 + w1 + n1) / (c + w1 + n1 + w2 + n2);
} }
@ -54,14 +46,11 @@ private:
* @param _set1 the first Pareto set * @param _set1 the first Pareto set
* @param _set2 the second Pareto set * @param _set2 the second Pareto set
*/ */
unsigned card_C (const std::vector < EOFitness > &_set1, unsigned card_C (const std::vector < ObjectiveVector > & _set1, const std::vector < ObjectiveVector > & _set2) {
const std::vector < EOFitness > &_set2) unsigned c=0;
{ for (unsigned i=0; i<_set1.size(); i++)
unsigned c = 0; for (unsigned j=0; j<_set2.size(); j++)
for (unsigned i = 0; i < _set1.size (); i++) if (_set1[i] == _set2[j]) {
for (unsigned j = 0; j < _set2.size (); j++)
if (_set1[i] == _set2[j])
{
c++; c++;
break; break;
} }
@ -73,14 +62,11 @@ private:
* @param _set1 the first Pareto set * @param _set1 the first Pareto set
* @param _set2 the second Pareto set * @param _set2 the second Pareto set
*/ */
unsigned card_W (const std::vector < EOFitness > &_set1, unsigned card_W (const std::vector < ObjectiveVector > & _set1, const std::vector < ObjectiveVector > & _set2) {
const std::vector < EOFitness > &_set2) unsigned w=0;
{ for (unsigned i=0; i<_set1.size(); i++)
unsigned w = 0; for (unsigned j=0; j<_set2.size(); j++)
for (unsigned i = 0; i < _set1.size (); i++) if (_set1[i].dominates(_set2[j])) {
for (unsigned j = 0; j < _set2.size (); j++)
if (_set1[i].dominates (_set2[j]))
{
w++; w++;
break; break;
} }
@ -92,20 +78,16 @@ private:
* @param _set1 the first Pareto set * @param _set1 the first Pareto set
* @param _set2 the second Pareto set * @param _set2 the second Pareto set
*/ */
unsigned card_N (const std::vector < EOFitness > &_set1, unsigned card_N (const std::vector < ObjectiveVector > & _set1, const std::vector < ObjectiveVector > & _set2) {
const std::vector < EOFitness > &_set2) unsigned n=0;
{ for (unsigned i=0; i<_set1.size(); i++) {
unsigned n = 0;
for (unsigned i = 0; i < _set1.size (); i++)
{
bool domin_rel = false; bool domin_rel = false;
for (unsigned j = 0; j < _set2.size (); j++) for (unsigned j=0; j<_set2.size(); j++)
if (_set1[i].dominates (_set2[j]) || _set2[j].dominates (_set1[i])) if (_set1[i].dominates(_set2[j]) || _set2[j].dominates(_set1 [i])) {
{
domin_rel = true; domin_rel = true;
break; break;
} }
if (!domin_rel) if (! domin_rel)
n++; n++;
} }
return n; return n;
@ -113,4 +95,4 @@ private:
}; };
#endif /*MOEOCONTRIBUTIONMETRIC_H_ */ #endif /*MOEOCONTRIBUTIONMETRIC_H_*/

178
trunk/paradiseo-moeo/src/metric/moeoEntropyMetric.h
View file

@ -2,7 +2,7 @@
//----------------------------------------------------------------------------- //-----------------------------------------------------------------------------
// moeoEntropyMetric.h // moeoEntropyMetric.h
// (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2006 // (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2007
/* /*
This library... This library...
@ -16,31 +16,23 @@
#include <metric/moeoMetric.h> #include <metric/moeoMetric.h>
/** /**
* The entropy gives an idea of the diversity of a Pareto set relatively to another Pareto set * The entropy gives an idea of the diversity of a Pareto set relatively to another
*
* (Basseur, Seynhaeve, Talbi: 'Design of Multi-objective Evolutionary Algorithms: Application to the Flow-shop Scheduling Problem', in Proc. of the 2002 Congress on Evolutionary Computation, IEEE Press, pp. 1155-1156) * (Basseur, Seynhaeve, Talbi: 'Design of Multi-objective Evolutionary Algorithms: Application to the Flow-shop Scheduling Problem', in Proc. of the 2002 Congress on Evolutionary Computation, IEEE Press, pp. 1155-1156)
*/ */
template < class EOT > class moeoEntropyMetric:public moeoVectorVsVectorBM < EOT, template < class ObjectiveVector >
double > class moeoEntropyMetric : public moeoVectorVsVectorBinaryMetric < ObjectiveVector, double >
{ {
public: public:
/**
* The fitness type of a solution
*/
typedef typename EOT::Fitness EOFitness;
/** /**
* Returns the entropy of the Pareto set '_set1' relatively to the Pareto set '_set2' * Returns the entropy of the Pareto set '_set1' relatively to the Pareto set '_set2'
* @param _set1 the first Pareto set * @param _set1 the first Pareto set
* @param _set2 the second Pareto set * @param _set2 the second Pareto set
*/ */
double operator () (const std::vector < EOFitness > &_set1, double operator()(const std::vector < ObjectiveVector > & _set1, const std::vector < ObjectiveVector > & _set2) {
const std::vector < EOFitness > &_set2)
{
// normalization // normalization
std::vector < EOFitness > set1 = _set1; std::vector< ObjectiveVector > set1 = _set1;
std::vector < EOFitness > set2 = _set2; std::vector< ObjectiveVector > set2= _set2;
removeDominated (set1); removeDominated (set1);
removeDominated (set2); removeDominated (set2);
prenormalize (set1); prenormalize (set1);
@ -48,33 +40,27 @@ public:
normalize (set2); normalize (set2);
// making of PO* // making of PO*
std::vector < EOFitness > star; // rotf :-) std::vector< ObjectiveVector > star; // rotf :-)
computeUnion (set1, set2, star); computeUnion (set1, set2, star);
removeDominated (star); removeDominated (star);
// making of PO1 U PO* // making of PO1 U PO*
std::vector < EOFitness > union_set1_star; // rotf again ... std::vector< ObjectiveVector > union_set1_star; // rotf again ...
computeUnion (set1, star, union_set1_star); computeUnion (set1, star, union_set1_star);
unsigned C = union_set1_star.size (); unsigned C = union_set1_star.size();
float omega = 0; float omega=0;
float entropy = 0; float entropy=0;
for (unsigned i = 0; i < C; i++) for (unsigned i=0 ; i<C ; i++) {
{ unsigned N_i = howManyInNicheOf (union_set1_star, union_set1_star[i], star.size());
unsigned N_i = unsigned n_i = howManyInNicheOf (set1, union_set1_star[i], star.size());
howManyInNicheOf (union_set1_star, union_set1_star[i], if (n_i > 0) {
star.size ());
unsigned n_i =
howManyInNicheOf (set1, union_set1_star[i], star.size ());
if (n_i > 0)
{
omega += 1.0 / N_i; omega += 1.0 / N_i;
entropy += entropy += (float) n_i / (N_i * C) * log (((float) n_i / C) / log (2.0));
(float) n_i / (N_i * C) * log (((float) n_i / C) / log (2.0));
} }
} }
entropy /= -log (omega); entropy /= - log (omega);
entropy *= log (2.0); entropy *= log (2.0);
return entropy; return entropy;
} }
@ -82,98 +68,110 @@ public:
private: private:
std::vector < double >vect_min_val; /** vector of min values */
std::vector < double >vect_max_val; std::vector<double> vect_min_val;
/** vector of max values */
std::vector<double> vect_max_val;
void removeDominated (std::vector < EOFitness > &_f)
{ /**
for (unsigned i = 0; i < _f.size (); i++) * Removes the dominated individuals contained in _f
{ * @param _f a Pareto set
*/
void removeDominated(std::vector < ObjectiveVector > & _f) {
for (unsigned i=0 ; i<_f.size(); i++) {
bool dom = false; bool dom = false;
for (unsigned j = 0; j < _f.size (); j++) for (unsigned j=0; j<_f.size(); j++)
if (i != j && _f[j].dominates (_f[i])) if (i != j && _f[j].dominates(_f[i])) {
{
dom = true; dom = true;
break; break;
} }
if (dom) if (dom) {
{ _f[i] = _f.back();
_f[i] = _f.back (); _f.pop_back();
_f.pop_back ();
i--; i--;
} }
} }
} }
void prenormalize (const std::vector < EOFitness > &_f)
{
vect_min_val.clear ();
vect_max_val.clear ();
for (unsigned char i = 0; i < EOFitness::fitness_traits::nObjectives (); /**
i++) * Prenormalization
{ * @param _f a Pareto set
float min_val = _f.front ()[i], max_val = min_val; */
for (unsigned j = 1; j < _f.size (); j++) void prenormalize (const std::vector< ObjectiveVector > & _f) {
{ vect_min_val.clear();
vect_max_val.clear();
for (unsigned char i=0 ; i<ObjectiveVector::nObjectives(); i++) {
float min_val = _f.front()[i], max_val = min_val;
for (unsigned j=1 ; j<_f.size(); j++) {
if (_f[j][i] < min_val) if (_f[j][i] < min_val)
min_val = _f[j][i]; min_val = _f[j][i];
if (_f[j][i] > max_val) if (_f[j][i]>max_val)
max_val = _f[j][i]; max_val = _f[j][i];
} }
vect_min_val.push_back (min_val); vect_min_val.push_back(min_val);
vect_max_val.push_back (max_val); vect_max_val.push_back (max_val);
} }
} }
void normalize (std::vector < EOFitness > &_f)
{ /**
for (unsigned i = 0; i < EOFitness::fitness_traits::nObjectives (); i++) * Normalization
for (unsigned j = 0; j < _f.size (); j++) * @param _f a Pareto set
_f[j][i] = */
(_f[j][i] - vect_min_val[i]) / (vect_max_val[i] - vect_min_val[i]); void normalize (std::vector< ObjectiveVector > & _f) {
for (unsigned i=0 ; i<ObjectiveVector::nObjectives(); i++)
for (unsigned j=0; j<_f.size(); j++)
_f[j][i] = (_f[j][i] - vect_min_val[i]) / (vect_max_val[i] - vect_min_val[i]);
} }
void computeUnion (const std::vector < EOFitness > &_f1,
const std::vector < EOFitness > &_f2, /**
std::vector < EOFitness > &_f) * Computation of the union of _f1 and _f2 in _f
{ * @param _f1 the first Pareto set
_f = _f1; * @param _f2 the second Pareto set
for (unsigned i = 0; i < _f2.size (); i++) * @param _f the final Pareto set
{ */
void computeUnion(const std::vector< ObjectiveVector > & _f1, const std::vector< ObjectiveVector > & _f2, std::vector< ObjectiveVector > & _f) {
_f = _f1 ;
for (unsigned i=0; i<_f2.size(); i++) {
bool b = false; bool b = false;
for (unsigned j = 0; j < _f1.size (); j++) for (unsigned j=0; j<_f1.size(); j ++)
if (_f1[j] == _f2[i]) if (_f1[j] == _f2[i]) {
{
b = true; b = true;
break; break;
} }
if (!b) if (! b)
_f.push_back (_f2[i]); _f.push_back(_f2[i]);
} }
} }
unsigned howManyInNicheOf (const std::vector < EOFitness > &_f,
const EOFitness & _s, unsigned _size) /**
{ * How many in niche
unsigned n = 0; */
for (unsigned i = 0; i < _f.size (); i++) unsigned howManyInNicheOf (const std::vector< ObjectiveVector > & _f, const ObjectiveVector & _s, unsigned _size) {
{ unsigned n=0;
if (euclidianDistance (_f[i], _s) < (_s.size () / (double) _size)) for (unsigned i=0 ; i<_f.size(); i++) {
if (euclidianDistance(_f[i], _s) < (_s.size() / (double) _size))
n++; n++;
} }
return n; return n;
} }
double euclidianDistance (const EOFitness & _set1, const EOFitness & _to,
unsigned _deg = 2) /**
{ * Euclidian distance
double dist = 0; */
for (unsigned i = 0; i < _set1.size (); i++) double euclidianDistance (const ObjectiveVector & _set1, const ObjectiveVector & _to, unsigned _deg = 2) {
dist += pow (fabs (_set1[i] - _to[i]), (int) _deg); double dist=0;
return pow (dist, 1.0 / _deg); for (unsigned i=0; i<_set1.size(); i++)
dist += pow(fabs(_set1[i] - _to[i]), (int)_deg);
return pow(dist, 1.0 / _deg);
} }
}; };
#endif /*MOEOENTROPYMETRIC_H_ */ #endif /*MOEOENTROPYMETRIC_H_*/

91
trunk/paradiseo-moeo/src/metric/moeoMetric.h
View file

@ -2,7 +2,7 @@
//----------------------------------------------------------------------------- //-----------------------------------------------------------------------------
// moeoMetric.h // moeoMetric.h
// (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2006 // (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2007
/* /*
This library... This library...
@ -16,91 +16,58 @@
#include <eoFunctor.h> #include <eoFunctor.h>
/** /**
* Base class for performance metrics (also called quality indicators) * Base class for performance metrics (also known as quality indicators).
*/ */
class moeoMetric:public eoFunctorBase class moeoMetric : public eoFunctorBase
{ {};
};
/** /**
* Base class for unary metrics * Base class for unary metrics.
*/ */
template < class A, class R > class moeoUM:public eoUF < A, R >, template < class A, class R >
public moeoMetric class moeoUnaryMetric : public eoUF < A, R >, public moeoMetric
{ {};
};
/** /**
* Base class for binary metrics * Base class for binary metrics.
*/ */
template < class A1, class A2, class R > class moeoBM:public eoBF < A1, A2, R >, template < class A1, class A2, class R >
public moeoMetric class moeoBinaryMetric : public eoBF < A1, A2, R >, public moeoMetric
{ {};
};
/** /**
* Base class for unary metrics dedicated to the performance evaluation of a single solution's Pareto fitness * Base class for unary metrics dedicated to the performance evaluation of a single solution's objective vector.
*/ */
template < class EOT, class R, class EOFitness = typename EOT::Fitness > class moeoSolutionUM:public moeoUM < template < class ObjectiveVector, class R >
const class moeoSolutionUnaryMetric : public moeoUnaryMetric < const ObjectiveVector &, R >
EOFitness &, {};
R >
{
};
/** /**
* Base class for unary metrics dedicated to the performance evaluation of a Pareto set (a vector of Pareto fitnesses) * Base class for unary metrics dedicated to the performance evaluation of a Pareto set (a vector of objective vectors)
*/ */
template < class EOT, class R, class EOFitness = typename EOT::Fitness > class moeoVectorUM:public moeoUM < template < class ObjectiveVector, class R >
const class moeoVectorUnaryMetric : public moeoUnaryMetric < const std::vector < ObjectiveVector > &, R >
std::vector < {};
EOFitness > &,
R >
{
};
/** /**
* Base class for binary metrics dedicated to the performance comparison between two solutions's Pareto fitnesses * Base class for binary metrics dedicated to the performance comparison between two solutions's objective vectors.
*/ */
template < class EOT, class R, class EOFitness = typename EOT::Fitness > class moeoSolutionVsSolutionBM:public moeoBM < template < class ObjectiveVector, class R >
const class moeoSolutionVsSolutionBinaryMetric : public moeoBinaryMetric < const ObjectiveVector &, const ObjectiveVector &, R >
EOFitness &, const {};
EOFitness &,
R >
{
};
/** /**
* Base class for binary metrics dedicated to the performance comparison between a Pareto set (a vector of Pareto fitnesses) and a single solution's Pareto fitness * Base class for binary metrics dedicated to the performance comparison between two Pareto sets (two vectors of objective vectors)
*/ */
template < class EOT, class R, class EOFitness = typename EOT::Fitness > class moeoVectorVsSolutionBM:public moeoBM < template < class ObjectiveVector, class R >
const class moeoVectorVsVectorBinaryMetric : public moeoBinaryMetric < const std::vector < ObjectiveVector > &, const std::vector < ObjectiveVector > &, R >
std::vector < {};
EOFitness > &, const
EOFitness &,
R >
{
};
/** #endif /*MOEOMETRIC_H_*/
* Base class for binary metrics dedicated to the performance comparison between two Pareto sets (two vectors of Pareto fitnesses)
*/
template < class EOT, class R, class EOFitness = typename EOT::Fitness > class moeoVectorVsVectorBM:public moeoBM <
const
std::vector <
EOFitness > &, const
std::vector <
EOFitness > &,
R >
{
};
#endif /*MOEOMETRIC_H_ */

271
trunk/paradiseo-moeo/src/metric/moeoNormalizedSolutionVsSolutionBinaryMetric.h Normal file
View file

@ -0,0 +1,271 @@
// -*- mode: c++; c-indent-level: 4; c++-member-init-indent: 8; comment-column: 35; -*-
//-----------------------------------------------------------------------------
// moeoNormalizedSolutionVsSolutionBinaryMetric.h
// (c) OPAC Team (LIFL), Dolphin Project (INRIA), 2007
/*
This library...
Contact: paradiseo-help@lists.gforge.inria.fr, http://paradiseo.gforge.inria.fr
*/
//-----------------------------------------------------------------------------
#ifndef MOEONORMALIZEDSOLUTIONVSSOLUTIONBINARYMETRIC_H_
#define MOEONORMALIZEDSOLUTIONVSSOLUTIONBINARYMETRIC_H_
#include <stdexcept>
#include <metric/moeoMetric.h>
/**
* Base class for binary metrics dedicated to the performance comparison between two solutions's objective vectors using normalized values.
* Then, indicator values lie in the interval [-1,1].
* Note that you have to set the bounds for every objective before using the operator().
*/
template < class ObjectiveVector, class R >
class moeoNormalizedSolutionVsSolutionBinaryMetric : public moeoSolutionVsSolutionBinaryMetric < ObjectiveVector, R >
{
public:
/**
* Default ctr for any moeoNormalizedSolutionVsSolutionBinaryMetric object
*/
moeoNormalizedSolutionVsSolutionBinaryMetric()
{
bounds.resize(ObjectiveVector::Traits::nObjectives());
}
/**
* Sets the lower bound (_min) and the upper bound (_max) for the objective _obj
* _min lower bound
* _max upper bound
* _obj the objective index
*/
void setup(double _min, double _max, unsigned _obj)
{
if (_min == _max)
{
_min -= tiny();
_max += tiny();
}
bounds[_obj] = eoRealInterval(_min, _max);
}
/**
* Sets the lower bound and the upper bound for the objective _obj using a eoRealInterval object
* _realInterval the eoRealInterval object
* _obj the objective index
*/
virtual void setup(eoRealInterval _realInterval, unsigned _obj)
{
bounds[_obj] = _realInterval;
}
/**
* Returns a very small value that can be used to avoid extreme cases (where the min bound == the max bound)
*/
static double tiny()
{
return 1e-6;
}
protected:
/** the bounds for every objective (bounds[i] = bounds for the objective i) */
std::vector < eoRealInterval > bounds;
};
/**
* Additive epsilon binary metric allowing to compare two objective vectors as proposed in
* Zitzler E., Thiele L., Laumanns M., Fonseca C. M., Grunert da Fonseca V.:
* Performance Assessment of Multiobjective Optimizers: An Analysis and Review. IEEE Transactions on Evolutionary Computation 7(2), pp.117–132 (2003).
*/
template < class ObjectiveVector >
class moeoAdditiveEpsilonBinaryMetric : public moeoNormalizedSolutionVsSolutionBinaryMetric < ObjectiveVector, double >
{
public:
/**
* Returns the minimal distance by which the objective vector _o1 must be translated in all objectives
* so that it weakly dominates the objective vector _o2
* @warning don't forget to set the bounds for every objective before the call of this function
* @param _o1 the first objective vector
* @param _o2 the second objective vector
*/
double operator()(const ObjectiveVector & _o1, const ObjectiveVector & _o2)
{
// computation of the epsilon value for the first objective
double result = epsilon(_o1, _o2, 0);
// computation of the epsilon value for the other objectives
double tmp;
for (unsigned i=1; i<ObjectiveVector::Traits::nObjectives(); i++)
{
tmp = epsilon(_o1, _o2, i);
result = std::max(result, tmp);
}
// returns the maximum epsilon value
return result;
}
private:
/** the bounds for every objective */
using moeoNormalizedSolutionVsSolutionBinaryMetric < ObjectiveVector, double > :: bounds;
/**
* Returns the epsilon value by which the objective vector _o1 must be translated in the objective _obj
* so that it dominates the objective vector _o2
* @param _o1 the first objective vector
* @param _o2 the second objective vector
* @param _obj the index of the objective
*/
double epsilon(const ObjectiveVector & _o1, const ObjectiveVector & _o2, const unsigned _obj)
{
double result;
// if the objective _obj have to be minimized
if (ObjectiveVector::Traits::minimizing(_obj))
{
// _o1[_obj] - _o2[_obj]
result = ( (_o1[_obj] - bounds[_obj].minimum()) / bounds[_obj].range() ) - ( (_o2[_obj] - bounds[_obj].minimum()) / bounds[_obj].range() );
}
// if the objective _obj have to be maximized
else
{
// _o2[_obj] - _o1[_obj]
result = ( (_o2[_obj] - bounds[_obj].minimum()) / bounds[_obj].range() ) - ( (_o1[_obj] - bounds[_obj].minimum()) / bounds[_obj].range() );
}
return result;
}
};
/**
* Hypervolume binary metric allowing to compare two objective vectors as proposed in
* Zitzler E., Künzli S.: Indicator-Based Selection in Multiobjective Search. In Parallel Problem Solving from Nature (PPSN VIII).
* Lecture Notes in Computer Science 3242, Springer, Birmingham, UK pp.832–842 (2004).
* This indicator is based on the hypervolume concept introduced in
* Zitzler, E., Thiele, L.: Multiobjective Optimization Using Evolutionary Algorithms - A Comparative Case Study.
* Parallel Problem Solving from Nature (PPSN-V), pp.292-301 (1998).
*/
template < class ObjectiveVector >
class moeoHypervolumeBinaryMetric : public moeoNormalizedSolutionVsSolutionBinaryMetric < ObjectiveVector, double >
{
public:
/**
* Ctor
* @param _rho value used to compute the reference point from the worst values for each objective (default : 1.1)
*/
moeoHypervolumeBinaryMetric(double _rho = 1.1) : rho(_rho)
{
// not-a-maximization problem check
for (unsigned i=0; i<ObjectiveVector::Traits::nObjectives(); i++)
{
if (ObjectiveVector::Traits::maximizing(i))
{
throw std::runtime_error("Hypervolume binary metric not yet implemented for a maximization problem in moeoHypervolumeBinaryMetric");
}
}
// consistency check
if (rho < 1)
{
cout << "Warning, value used to compute the reference point rho for the hypervolume calculation must not be smaller than 1" << endl;
cout << "Adjusted to 1" << endl;
rho = 1;
}
}
/**
* Returns the volume of the space that is dominated by _o2 but not by _o1 with respect to a reference point computed using rho.
* @warning don't forget to set the bounds for every objective before the call of this function
* @param _o1 the first objective vector
* @param _o2 the second objective vector
*/
double operator()(const ObjectiveVector & _o1, const ObjectiveVector & _o2)
{
double result;
// if _o1 dominates _o2
if ( paretoComparator(_o1,_o2) )
{
result = - hypervolume(_o1, _o2, ObjectiveVector::Traits::nObjectives()-1);
}
else
{
result = hypervolume(_o2, _o1, ObjectiveVector::Traits::nObjectives()-1);
}
return result;
}
private:
/** value used to compute the reference point from the worst values for each objective */
double rho;
/** the bounds for every objective */
using moeoNormalizedSolutionVsSolutionBinaryMetric < ObjectiveVector, double > :: bounds;
/** Functor to compare two objective vectors according to Pareto dominance relation */
moeoParetoObjectiveVectorComparator < ObjectiveVector > paretoComparator;
/**
* Returns the volume of the space that is dominated by _o2 but not by _o1 with respect to a reference point computed using rho for the objective _obj.
* @param _o1 the first objective vector
* @param _o2 the second objective vector
* @param _obj the objective index
* @param _flag used for iteration, if _flag=true _o2 is not talen into account (default : false)
*/
double hypervolume(const ObjectiveVector & _o1, const ObjectiveVector & _o2, const unsigned _obj, const bool _flag = false)
{
double result;
double range = rho * bounds[_obj].range();
double max = bounds[_obj].minimum() + range;
// value of _1 for the objective _obj
double v1 = _o1[_obj];
// value of _2 for the objective _obj (if _flag=true, v2=max)
double v2;
if (_flag)
{
v2 = max;
}
else
{
v2 = _o2[_obj];
}
// computation of the volume
if (_obj == 0)
{
if (v1 < v2)
{
result = (v2 - v1) / range;
}
else
{
result = 0;
}
}
else
{
if (v1 < v2)
{
result = ( hypervolume(_o1, _o2, _obj-1, true) * (v2 - v1) / range ) + ( hypervolume(_o1, _o2, _obj-1) * (max - v2) / range );
}
else
{
result = hypervolume(_o1, _o2, _obj-1) * (max - v2) / range;
}
}
return result;
}
};
#endif /*MOEONORMALIZEDSOLUTIONVSSOLUTIONBINARYMETRIC_H_*/