files for trikisa

This commit is contained in:
LPTK 2013-06-12 15:19:32 +02:00
commit 23a7482f1c
8 changed files with 812 additions and 0 deletions

View file

@ -0,0 +1,119 @@
/*
<moBestFitnessStat.h>
Copyright (C) DOLPHIN Project-Team, INRIA Lille - Nord Europe, 2006-2010
Sébastien Verel, Arnaud Liefooghe, Jérémie Humeau
This software is governed by the CeCILL license under French law and
abiding by the rules of distribution of free software. You can use,
modify and/ or redistribute the software under the terms of the CeCILL
license as circulated by CEA, CNRS and INRIA at the following URL
"http://www.cecill.info".
As a counterpart to the access to the source code and rights to copy,
modify and redistribute granted by the license, users are provided only
with a limited warranty and the software's author, the holder of the
economic rights, and the successive licensors have only limited liability.
In this respect, the user's attention is drawn to the risks associated
with loading, using, modifying and/or developing or reproducing the
software by the user in light of its specific status of free software,
that may mean that it is complicated to manipulate, and that also
therefore means that it is reserved for developers and experienced
professionals having in-depth computer knowledge. Users are therefore
encouraged to load and test the software's suitability as regards their
requirements in conditions enabling the security of their systems and/or
data to be ensured and, more generally, to use and operate it in the
same conditions as regards security.
The fact that you are presently reading this means that you have had
knowledge of the CeCILL license and that you accept its terms.
ParadisEO WebSite : http://paradiseo.gforge.inria.fr
Contact: paradiseo-help@lists.gforge.inria.fr
*/
#ifndef moFitnessMomentsStat_h
#define moFitnessMomentsStat_h
#include <utility>
#include <continuator/moStat.h>
/**
* Statistic that saves the standard deviation of the fitness of the solutions during the search
*/
template <class EOT>
//class moFitnessMomentsStat : public moStat<EOT, std::pair<typename EOT::Fitness,typename EOT::Fitness> >
class moFitnessMomentsStat : public moStat<EOT, std::pair<double, double> >
{
public :
typedef typename EOT::Fitness Fitness;
//typedef std::pair<typename EOT::Fitness,typename EOT::Fitness> Pair;
typedef std::pair<double, double> Pair;
using moStat<EOT, Pair >::value;
/**
* Default Constructor
* @param _reInitSol when true the best so far is reinitialized
*/
moFitnessMomentsStat(bool _reInitSol = true)
: moStat<EOT, Pair>(Pair(Fitness(), 0.0), "moments (average and stdev)"),
reInitSol(_reInitSol), firstTime(true),
nbSolutionsEncountered(0), currentAvg(0), currentVar(0)
{ }
/**
* Initialization of the best solution on the first one
* @param _sol the first solution
*/
virtual void init(EOT & _sol) {
if (reInitSol || firstTime)
{
value() = Pair(0.0,0.0);
nbSolutionsEncountered = currentAvg = currentVar = 0;
firstTime = false;
}
/*else if (firstTime)
{
value() = 0.0;
firstTime = false;
}*/
operator()(_sol);
}
/**
* Update the best solution
* @param _sol the current solution
*/
virtual void operator()(EOT & _sol) {
++nbSolutionsEncountered;
double x = _sol.fitness();
double oldAvg = currentAvg;
currentAvg = oldAvg + (x - oldAvg)/nbSolutionsEncountered;
if (nbSolutionsEncountered > 1) // <- not really necessary
{
//value() = (value()/nbSolutionsEncountered + _sol.fitness())/(nbSolutionsEncountered+1);
double oldVar = currentVar;
currentVar = oldVar + (x - oldAvg) * (x - currentAvg);
value() = Pair(currentAvg, currentVar/nbSolutionsEncountered);
}
}
/**
* @return name of the class
*/
virtual std::string className(void) const {
return "moFitnessVarianceStat";
}
protected:
bool reInitSol;
bool firstTime;
double
nbSolutionsEncountered
, currentAvg
, currentVar // actually the var * n
;
};
#endif // moFitnessMomentsStat_h

View file

@ -0,0 +1,143 @@
#ifndef __moStdDevEstimator_h__
#define __moStdDevEstimator_h__
#include <eo>
#include <mo>
#include <moFitnessVarianceStat.h> // TODO rm
#include <iostream> // TODO rm
// TODO make tests
template< class T >
class eoOptional {
public:
static const eoOptional<T> null; // = eoOptional<T>();
eoOptional (T& init)
: _val(&init)
{ }
bool hasValue() const
{
return _val != NULL;
}
T& get () const
{
if (!hasValue())
throw std::runtime_error("Cannot get a reference from a eoOptional wrapper with no value");
return *_val;
}
protected:
eoOptional ()
: _val(NULL)
{ }
private:
T* _val;
};
template< class T >
const eoOptional<T> eoOptional<T>::null = eoOptional<T>();
template< class EOT, class Neighbor >
class moStdDevEstimator : public eoUF<EOT&, double>
{
public:
/**
* General constructor for the estimator
* @param continuator a user-defined continuator
* @param neighborhood the neighborhood
* @param fullEval the full evaluation function
* @param eval neighbor's evaluation function
* @param walker a local search algorithm
*/
moStdDevEstimator<EOT,Neighbor> (
moContinuator<Neighbor>& continuator,
moNeighborhood < Neighbor > & neighborhood,
eoEvalFunc<EOT>& fullEval,
/* The following should be read:
moEval<Neighbor> >& eval = _default_eval
* (which is not possible to achieve as is in C++) */
const eoOptional< moEval<Neighbor> >& eval = eoOptional< moEval<Neighbor> >::null,
const eoOptional< moLocalSearch<Neighbor> >& walker = eoOptional< moLocalSearch<Neighbor> >::null
)
: _default_eval ( fullEval ),
_eval(eval.hasValue()? eval.get(): _default_eval),
_default_continuator( 0 ),
_continuator( _continuator ),
_checkpoint( _continuator ),
_default_walker( neighborhood, fullEval, _eval, _checkpoint ),
_walker( walker.hasValue()? walker.get(): _default_walker )
{
_checkpoint.add( _varStat );
}
/**
* Simpler constructor for the estimator
* @param max_iters the number of steps the default moIterContinuator should perform
* @param neighborhood the neighborhood
* @param fullEval the full evaluation function
* @param eval neighbor's evaluation function
* @param walker a local search algorithm
*/
moStdDevEstimator<EOT,Neighbor> (
unsigned int max_iters,
moNeighborhood < Neighbor > & neighborhood,
eoEvalFunc<EOT>& fullEval,
const eoOptional< moEval<Neighbor> >& eval = eoOptional< moEval<Neighbor> >::null,
const eoOptional< moLocalSearch<Neighbor> >& walker = eoOptional< moLocalSearch<Neighbor> >::null
)
: _default_eval ( fullEval ),
_eval(eval.hasValue()? eval.get(): _default_eval),
_default_continuator( max_iters, false ),
_continuator( _default_continuator ),
_checkpoint( _continuator ),
_default_walker( neighborhood, fullEval, _eval, _checkpoint ),
_walker( walker.hasValue()? walker.get(): _default_walker )
{
_checkpoint.add( _varStat );
}
/**
* Evaluates the estimator with the walker algorithm and returns the standard deviation
* @param solution the solution from where to start the walk
*/
double operator()( EOT & solution )
{
_walker(solution);
return sqrt(_varStat.value());
}
/**
* @return the class name
*/
virtual std::string className(void) const {
return "moStdDevEstimator";
}
private:
moFullEvalByCopy <Neighbor> _default_eval;
moEval<Neighbor>& _eval;
moIterContinuator <Neighbor> _default_continuator;
moContinuator <Neighbor>& _continuator;
moCheckpoint <Neighbor> _checkpoint;
moRandomWalk <Neighbor> _default_walker;
moLocalSearch <Neighbor> _walker;
moFitnessVarianceStat<EOT> _varStat;
};
#endif // __moStdDevEstimator_h__

View file

@ -0,0 +1,363 @@
/*
<moTrikiCoolingSchedule.h>
Copyright (C) DOLPHIN Project-Team, INRIA Futurs, 2006-2008
(C) OPAC Team, LIFL, 2002-2008
Sébastien Cahon, Jean-Charles Boisson (Jean-Charles.Boisson@lifl.fr)
This software is governed by the CeCILL license under French law and
abiding by the rules of distribution of free software. You can use,
modify and/ or redistribute the software under the terms of the CeCILL
license as circulated by CEA, CNRS and INRIA at the following URL
"http://www.cecill.info".
As a counterpart to the access to the source code and rights to copy,
modify and redistribute granted by the license, users are provided only
with a limited warranty and the software's author, the holder of the
economic rights, and the successive licensors have only limited liability.
In this respect, the user's attention is drawn to the risks associated
with loading, using, modifying and/or developing or reproducing the
software by the user in light of its specific status of free software,
that may mean that it is complicated to manipulate, and that also
therefore means that it is reserved for developers and experienced
professionals having in-depth computer knowledge. Users are therefore
encouraged to load and test the software's suitability as regards their
requirements in conditions enabling the security of their systems and/or
data to be ensured and, more generally, to use and operate it in the
same conditions as regards security.
The fact that you are presently reading this means that you have had
knowledge of the CeCILL license and that you accept its terms.
ParadisEO WebSite : http://paradiseo.gforge.inria.fr
Contact: paradiseo-help@lists.gforge.inria.fr
*/
#ifndef _moTrikiCoolingSchedule_h
#define _moTrikiCoolingSchedule_h
#include <coolingSchedule/moCoolingSchedule.h>
#include <continuator/moNeighborhoodStat.h>
#include <continuator/moStdFitnessNeighborStat.h>
#include <neighborhood/moNeighborhood.h>
#include <continuator/moStat.h>
#include <moFitnessMomentsStat.h>
/*
#include <continuator/moStat.h>
#include <explorer/moNeighborhoodExplorer.h>
#include <comparator/moNeighborComparator.h>
#include <comparator/moSolNeighborComparator.h>
#include <neighborhood/moNeighborhood.h>
*/
#include <iostream>
using namespace std;
//!
/*!
*/
template< class EOT, class Neighbor > //, class Neighborhood >
class moTrikiCoolingSchedule: public moCoolingSchedule< EOT >
{
public:
//typedef typename Neighbor::EOT EOT ;
typedef moNeighborhood<Neighbor> Neighborhood ;
//! Constructor
/*!
*/
moTrikiCoolingSchedule (Neighborhood& _neighborhood, moEval<Neighbor>& _eval, double _initTemp)
: initTemp(_initTemp),
mu2(10), // mu2 typically belongs to [1; 20]
K1(2), // K1 in [1; 4], the number of chains without reaching equilibrium before we raise the temperature
K2(5), // ???
lambda1(2), // the increase in temperature, typically in [1.5; 4]
lambda2(.7), // lambda2 in [0.5; 0.99]
mu1(10), // target decrease in cost factor, in [2; 20]
xi(1.05), // xi typically belongs to [1; 1.1]
max_accepted(50), // depends on pb/neighborhood
max_generated(100), // depends on pb/neighborhood
theta(10), // theta is typically set to 10
statIsInitialized(false),
outf("out.data")
{ }
moTrikiCoolingSchedule (
Neighborhood& _neighborhood, moEval<Neighbor>& _eval, double _initTemp,
double _max_accepted,
double _max_generated
)
: initTemp(_initTemp),
mu2(10), // mu2 typically belongs to [1; 20]
K1(2), // K1 in [1; 4], the number of chains without reaching equilibrium before we raise the temperature
K2(5), // ???
lambda1(2), // the increase in temperature, typically in [1.5; 4]
lambda2(.7), // lambda2 in [0.5; 0.99]
mu1(10), // target decrease in cost factor, in [2; 20]
xi(1.05), // xi typically belongs to [1; 1.1]
max_accepted(_max_accepted), // depends on pb/neighborhood
max_generated(_max_generated), // depends on pb/neighborhood
theta(10), // theta is typically set to 10
statIsInitialized(false),
outf("out.data")
{ }
/**
* Initial temperature
* @param _solution initial solution
*/
double init(EOT & _solution) {
accepted = generated = costs_sum = 0;
negative_temp = equilibrium_not_reached = frozen = 0;
reinitializing = false;
terminated = false;
statIsInitialized = false;
///
cout << "INIT T=" << initTemp << endl;
///
//outf.open("out");
//outf << "ok";
//outf.close();
return initTemp;
}
/**
* update the temperature by a factor
* @param _temp current temperature to update
* @param _acceptedMove true when the move is accepted, false otherwise
*/
void update(double& _temp, bool _acceptedMove, EOT & _solution) {
//cout << _temp << " g " << generated << endl;
generated++;
if (_acceptedMove)
{
accepted++;
//costs_sum += _solution.fitness();
//varStat(_solution);
if (statIsInitialized)
momStat(_solution);
else momStat.init(_solution), statIsInitialized = true;
//cout << _solution.fitness() << " avgCost=" << momStat.value().first << endl;
}
if (accepted > max_accepted || generated > max_generated) {
if (accepted == 0) // ADDED! Otherwise the computed std dev is null; we're probably at equilibrium
{
///
cout << "Stopping: no accepted solution" << endl;
///
terminated = true;
return;
}
///
cout << (accepted > max_accepted? "MAXACC ": "MAXGEN ");
///
//double avgCost = costs_sum/(double)accepted;
//double stdDev = sqrt(varStat.value()); // WARNING: IT'S NO MORE THE AVG COST, NOW IT'S THE STD DEV!
//double variance = varStat.value();
double avgCost = momStat.value().first;
double variance = momStat.value().second;
double stdDev = sqrt(variance);
double sigma = stdDev;
double delta = sigma/mu2;
//outf << avgCost << endl;
//outf << _temp << endl;
outf << prevAvgCost-delta << endl;
accepted = generated = costs_sum = 0;
//varStat.init(_solution);//TODON use next chain's first sol
//momStat.init(_solution);//TODONE use next chain's first sol
statIsInitialized = false;
///
cout << "T=" << _temp << " avgCost=" << avgCost << " stdDev=" << stdDev << " currCost=" << _solution.fitness() << endl;
///
double alpha;
double oldprevAvgCost = prevAvgCost;
///
cout << "negTemp: " << negative_temp << " / " << K2 << endl;
///
if (negative_temp < K2)
{
if (!reinitializing)
{
///
if (avgCost/(prevAvgCost-delta) > xi) cout << "/!\\ eq not reached!" << endl;
///
if (avgCost/(prevAvgCost-delta) > xi)
equilibrium_not_reached++;
else equilibrium_not_reached = 0;
}
if (equilibrium_not_reached > K1)
{
///
cout << "/!\\ Reinitializing (eq not reached)" << endl;
///
reinitializing = true;
alpha = lambda1;
delta = sigma/mu1;
equilibrium_not_reached = 0; // ADDED! Otherwise the algo gets trapped here!
}
else if (_temp*delta/(sigma*sigma) >= 1)
{
///
cout << "/!\\ neg temp!" << endl;
///
negative_temp++;
reinitializing = true;
if (negative_temp < K2)
{
alpha = lambda1;
delta = sigma/mu1;
} else
alpha = lambda2;
}
// First interpretation of the pseudocode indentation: (seems obviously false because it makes the above code unreachable)
/*
}
else
{
cout << "ccc" << endl;
reinitializing = false;
prevAvgCost = avgCost;
alpha = 1-_temp*delta/(sigma*sigma);
}
*/
// Second interpretation of the pseudocode indentation:
else
{
///
cout << "[normal decrease]" << endl;
///
reinitializing = false;
prevAvgCost = avgCost;
//alpha = 1-_temp*delta/(sigma*sigma);
alpha = 1-_temp*delta/variance;
//alpha = (sigma==0? 1: 1-_temp*delta/(sigma*sigma)); // ADDED! but removed
if (sigma == 0) // ADDED! When std dev is null, the solution is probably at eq, and the algo can't go on anyways
terminated = true, cout << "Stopping: null std dev" << endl;
}
}
// FIXME: else what? alpha=?
///
cout << "*=" << alpha << endl;
///
_temp *= alpha;
// Never seems to be used
if (avgCost == oldprevAvgCost) // use a neighborhood to approximate double equality?
frozen++;
else frozen = 0;
//exit(0);
//cin.get();
}
}
//! Function which proceeds to the cooling
/*!
*/
bool operator() (double temperature)
{
///
if (terminated) cout << "TERMINATED" << endl;
///
return frozen < theta
&& !terminated ; // ADDED! because 'frozen' doesn't terminate anything
}
private:
//moNeighborhoodStat<Neighbor> nhStat;
//moStdFitnessNeighborStat<Neighbor> stdDevStat;
const double
// parameters of the algorithm
//currentTemp,
initTemp,
//ratio,
//threshold,
mu2, // mu2 typically belongs to [1; 20]
K1, // K1 in [1; 4], the number of chains without reaching equilibrium before we raise the temperature
K2,
lambda1, // the increase in temperature, typically in [1.5; 4]
lambda2, // lambda2 in [0.5; 0.99]
mu1, // target decrease in cost factor, in [2; 20]
xi // xi typically belongs to [1; 1.1]
// private variables
;
double
stdDev,
prevAvgCost,
expectedDecreaseInCost, // delta
costs_sum
;
const int
max_accepted,
max_generated,
theta // theta is typically set to 10
;
int
accepted,
generated,
equilibrium_not_reached,
negative_temp,
frozen
;
bool reinitializing, terminated;
//moFitnessVarianceStat<EOT> varStat;
moFitnessMomentsStat<EOT> momStat;
bool statIsInitialized;
ofstream outf;
};
#endif

View file

@ -0,0 +1,87 @@
#ifndef __moRealNeighbor_h__
#define __moRealNeighbor_h__
#include <mo>
#include <eo>
#include <edo>
//! A neighbor as produced by a moRealNeighborhood
/*!
* In a real neighborhood, the move is just a translation vector, of the same type than a solution.
*/
template <class EOT, class Fitness=typename EOT::Fitness>
class moRealNeighbor : public moNeighbor<EOT, Fitness>
{
protected:
//! The move to be applied
EOT _translation;
edoBounder<EOT> * _bounder;
public:
moRealNeighbor<EOT,Fitness>() : _bounder( NULL ) { }
//! Returns the solution attached to this neighbor
EOT translation() { return _translation; }
//! Set the translation
void translation( EOT translation ) { _translation = translation; }
void bounder( edoBounder<EOT> * bounder ) { _bounder = bounder; }
/**
* Assignment operator
* @param _neighbor the neighbor to assign
* @return a neighbor equal to the other
*/
virtual moNeighbor<EOT, Fitness>& operator=(const moNeighbor<EOT, Fitness>& _neighbor) {
fitness( _neighbor.fitness() );
return (*this);
}
/*!
* Move a solution to the solution of this neighbor
* @param _solution the related solution
*/
virtual void move(EOT & _solution)
{
assert( _solution.size() == _translation.size() );
for( unsigned int i=0, size= _solution.size(); i<size; ++i ) {
_solution[i] += _translation[i];
}
if( _bounder != NULL ) {
(*_bounder)( _solution );
}
_solution.invalidate();
}
/**
* Test equality between two neighbors
* @param _neighbor a neighbor
* @return if _neighbor and this one are equals
*/
virtual bool equals(moRealNeighbor<EOT>& _neighbor) {
return _neighbor.translation() == _translation;
}
/**
* Return the class Name
* @return the class name as a std::string
*/
virtual std::string className() const {
return "moRealNeighbor";
}
};
#endif // __moRealNeighbor_h__

View file

@ -0,0 +1,78 @@
#ifndef __moRealNeighborhood_h__
#define __moRealNeighborhood_h__
#include <mo>
#include <eoFunctor.h> // FIXME: Why don't we use eoFunctorBase on the mother classes
#include "moRealNeighbor.h"
template<class Distrib, class Neighbor>
class moRealNeighborhood : public moRndNeighborhood< Neighbor >, public eoFunctorBase
{
public:
typedef typename Distrib::EOType EOT;
protected:
Distrib & _distrib;
edoSampler<Distrib> & _sampler;
edoBounder<EOT> & _bounder;
public:
moRealNeighborhood( Distrib & distrib, edoSampler<Distrib> & sampler, edoBounder<EOT> & bounder ) : _distrib(distrib), _sampler(sampler), _bounder(bounder) {}
/**
* It alway remains at least a solution in an infinite neighborhood
* @param _solution the related solution
* @return true
*/
virtual bool hasNeighbor(EOT &)
{
return true;
}
/**
* Draw the next neighbor
* @param _solution the solution to explore
* @param _current the next neighbor
*/
virtual void next(EOT &, Neighbor & _current)
{
_current.bounder( &_bounder );
// Draw a translation in the distrib, using the sampler
_current.translation( _sampler( _distrib ) );
}
/**
* Initialization of the neighborhood
* @param _solution the solution to explore
* @param _current the first neighbor
*/
virtual void init(EOT & _solution, Neighbor & _current)
{
// there is no difference between an init and a random draw
next( _solution, _current );
}
/**
* There is always a solution in an infinite neighborhood
* @param _solution the solution to explore
* @return true
*/
virtual bool cont(EOT &)
{
return true;
}
/**
* Return the class Name
* @return the class name as a std::string
*/
virtual std::string className() const {
return "moRealNeighborhood";
}
};
#endif // __moRealNeighborhood_h__

22
mo/src/trikisa Normal file
View file

@ -0,0 +1,22 @@
// (c) Thales group, 2010
/*
Authors:
Johann Dreo <johann.dreo@thalesgroup.com>
Caner Candan <caner.candan@thalesgroup.com>
*/
#ifndef _trikisa_
#define _trikisa_
#include "moRealNeighbor.h"
#include "moRealNeighborhood.h"
#include "moStdDevEstimator.h"
#include "moTrikiCoolingSchedule.h"
#include "moFitnessVarianceStat.h" // TODO rm
#include "moFitnessMomentsStat.h"
#endif // !_trikisa_
// Local Variables:
// mode: C++
// End:

0
mo/test/t-moTriki.cpp Normal file
View file

View file