MPI MultiStart: using SGA as example and functors for seed generation, reinitialization of pop, algorithm reset.

This commit is contained in:
Benjamin Bouvier 2012-07-26 11:58:42 +02:00
commit 0c1fc2ce99

View file

@ -1,76 +1,79 @@
# include <mpi/eoMpi.h>
using namespace eo::mpi;
#include <stdexcept>
#include <iostream>
#include <sstream>
#include <eo>
#include <es.h>
/***************************** EASY PSO STUFF ********************************/
//-----------------------------------------------------------------------------
typedef eoMinimizingFitness ParticleFitness;
//-----------------------------------------------------------------------------
class SerializableParticle : public eoRealParticle< ParticleFitness >, public eoserial::Persistent
// Use functions from namespace std
using namespace std;
class SerializableEOReal: public eoReal<double>, public eoserial::Persistent
{
public:
public:
SerializableParticle(unsigned size = 0, double positions = 0.0,double velocities = 0.0,double bestPositions = 0.0): eoRealParticle< ParticleFitness > (size, positions,velocities,bestPositions) {}
SerializableEOReal(unsigned size = 0, double value = 0.0) :
eoReal<double>(size, value)
{
}
void unpack( const eoserial::Object* obj )
void unpack( const eoserial::Object* obj )
{
this->clear();
eoserial::unpackArray
< std::vector<double>, eoserial::Array::UnpackAlgorithm >
( *obj, "vector", *this );
bool invalidFitness;
eoserial::unpack( *obj, "invalid_fitness", invalidFitness );
if( invalidFitness )
{
this->clear();
eoserial::unpackArray
< std::vector<double>, eoserial::Array::UnpackAlgorithm >
( *obj, "vector", *this );
this->invalidate();
} else
{
double f;
eoserial::unpack( *obj, "fitness", f );
this->fitness( f );
}
}
this->bestPositions.clear();
eoserial::unpackArray
< std::vector<double>, eoserial::Array::UnpackAlgorithm >
( *obj, "best_positions", this->bestPositions );
eoserial::Object* pack( void ) const
{
eoserial::Object* obj = new eoserial::Object;
obj->add( "vector", eoserial::makeArray< std::vector<double>, eoserial::MakeAlgorithm >( *this ) );
this->velocities.clear();
eoserial::unpackArray
< std::vector<double>, eoserial::Array::UnpackAlgorithm >
( *obj, "velocities", this->velocities );
bool invalidFitness;
eoserial::unpack( *obj, "invalid_fitness", invalidFitness );
if( invalidFitness )
{
this->invalidate();
} else
{
ParticleFitness f;
eoserial::unpack( *obj, "fitness", f );
this->fitness( f );
}
}
eoserial::Object* pack( void ) const
{
eoserial::Object* obj = new eoserial::Object;
obj->add( "vector", eoserial::makeArray< std::vector<double>, eoserial::MakeAlgorithm >( *this ) );
obj->add( "best_positions", eoserial::makeArray< std::vector<double>, eoserial::MakeAlgorithm >( this->bestPositions ) );
obj->add( "velocities", eoserial::makeArray< std::vector<double>, eoserial::MakeAlgorithm>( this->velocities ) );
bool invalidFitness = this->invalid();
obj->add( "invalid_fitness", eoserial::make( invalidFitness ) );
if( !invalidFitness )
{
obj->add( "fitness", eoserial::make( this->fitness() ) );
}
return obj;
bool invalidFitness = this->invalid();
obj->add( "invalid_fitness", eoserial::make( invalidFitness ) );
if( !invalidFitness )
{
obj->add( "fitness", eoserial::make( this->fitness() ) );
}
return obj;
}
};
typedef SerializableParticle Particle;
//-----------------------------------------------------------------------------
// the objective function
double real_value (const Particle & _particle)
// REPRESENTATION
//-----------------------------------------------------------------------------
// define your individuals
typedef SerializableEOReal Indi;
typedef double IndiFitness;
// EVAL
//-----------------------------------------------------------------------------
// a simple fitness function that computes the euclidian norm of a real vector
// @param _indi A real-valued individual
double real_value(const Indi & _indi)
{
double sum = 0;
for (unsigned i = 0; i < _particle.size ()-1; i++)
sum += pow(_particle[i],2);
return (sum);
double sum = 0;
for (unsigned i = 0; i < _indi.size(); i++)
sum += _indi[i]*_indi[i];
return (-sum); // maximizing only
}
/************************** PARALLELIZATION JOB *******************************/
@ -113,10 +116,12 @@ struct SerializableBasicType : public eoserial::Persistent
template< class EOT, class FitT >
struct MultiStartData
{
MultiStartData( mpi::communicator& _comm, eoAlgo<EOT>& _algo, int _masterRank, eoInit<EOT>* _init = 0 )
typedef eoF<void> ResetAlgo;
MultiStartData( mpi::communicator& _comm, eoAlgo<EOT>& _algo, int _masterRank, ResetAlgo & _resetAlgo )
:
runs( 0 ), firstIndividual( true ), bestFitness(), pop(),
comm( _comm ), algo( _algo ), masterRank( _masterRank ), init( _init )
comm( _comm ), algo( _algo ), masterRank( _masterRank ), resetAlgo( _resetAlgo )
{
// empty
}
@ -131,7 +136,7 @@ struct MultiStartData
// static parameters
mpi::communicator& comm;
eoAlgo<EOT>& algo;
eoInit<EOT>* init;
ResetAlgo& resetAlgo;
int masterRank;
};
@ -183,6 +188,13 @@ class ProcessTaskMultiStart : public ProcessTaskFunction< MultiStartData< EOT, F
void operator()()
{
// DEBUG
//static int i = 0;
//std::cout << Node::comm().rank() << "-" << i++ << " random: " << eo::rng.rand() << std::endl;
// std::cout << "POP(" << _data->pop.size() << ") : " << _data->pop << std::endl;
_data->resetAlgo();
_data->algo( _data->pop );
_data->comm.send( _data->masterRank, 1, _data->pop.best_element() );
}
@ -205,10 +217,22 @@ class MultiStartStore : public JobStore< MultiStartData< EOT, FitT > >
{
public:
MultiStartStore( eoAlgo<EOT> & algo, int masterRank, const eoPop< EOT > & pop, eoInit<EOT>* init = 0 )
: _data( Node::comm(), algo, masterRank, init ),
_pop( pop ),
_firstPopInit( true )
typedef typename MultiStartData<EOT,FitT>::ResetAlgo ResetAlgo;
typedef eoUF< eoPop<EOT>&, void > ReinitJob;
typedef eoUF< int, std::vector<int> > GetSeeds;
MultiStartStore(
eoAlgo<EOT> & algo,
int masterRank,
// eoInit<EOT>* init = 0
ReinitJob & reinitJob,
ResetAlgo & resetAlgo,
GetSeeds & getSeeds
)
: _data( Node::comm(), algo, masterRank, resetAlgo ),
_masterRank( masterRank ),
_getSeeds( getSeeds ),
_reinitJob( reinitJob )
{
this->_iff = new IsFinishedMultiStart< EOT, FitT >;
this->_iff->needDelete(true);
@ -220,20 +244,39 @@ class MultiStartStore : public JobStore< MultiStartData< EOT, FitT > >
this->_ptf->needDelete(true);
}
void init( int runs )
void init( const std::vector<int>& workers, int runs )
{
int nbWorkers = workers.size();
_reinitJob( _data.pop );
_data.runs = runs;
if( _data.init )
std::vector< int > seeds = _getSeeds( nbWorkers );
if( Node::comm().rank() == _masterRank )
{
_data.pop = eoPop<EOT>( _pop.size(), *_data.init );
} else if( _firstPopInit )
{
_data.pop = _pop;
}
_firstPopInit = false;
if( seeds.size() < nbWorkers )
{
// TODO
// get multiples of the current seed?
// generate seeds?
for( int i = 1; seeds.size() < nbWorkers ; ++i )
{
seeds.push_back( i );
}
}
_data.firstIndividual = true;
for( int i = 0 ; i < nbWorkers ; ++i )
{
int wrkRank = workers[i];
Node::comm().send( wrkRank, 1, seeds[ i ] );
}
} else
{
int seed;
Node::comm().recv( _masterRank, 1, seed );
std::cout << Node::comm().rank() << "- Seed: " << seed << std::endl;
eo::rng.reseed( seed );
}
}
MultiStartData<EOT, FitT>* data()
@ -243,12 +286,14 @@ class MultiStartStore : public JobStore< MultiStartData< EOT, FitT > >
private:
MultiStartData< EOT, FitT > _data;
const eoPop< EOT >& _pop;
bool _firstPopInit;
GetSeeds & _getSeeds;
ReinitJob & _reinitJob;
int _masterRank;
};
template< class EOT, class FitT >
class MultiStart : public MultiJob< MultiStartData< EOT, FitT > >
class MultiStart : public OneShotJob< MultiStartData< EOT, FitT > >
{
public:
@ -258,38 +303,9 @@ class MultiStart : public MultiJob< MultiStartData< EOT, FitT > >
// dynamic parameters
int runs,
const std::vector<int>& seeds = std::vector<int>() ) :
MultiJob< MultiStartData< EOT, FitT > >( algo, masterRank, store )
OneShotJob< MultiStartData< EOT, FitT > >( algo, masterRank, store )
{
store.init( runs );
if( this->isMaster() )
{
int nbWorkers = algo.availableWorkers();
std::vector<int> realSeeds = seeds;
if( realSeeds.size() < nbWorkers )
{
// TODO
// get multiples of the current seed?
// generate seeds?
for( int i = 1; realSeeds.size() < nbWorkers ; ++i )
{
realSeeds.push_back( i );
}
}
std::vector<int> idles = algo.idles();
for( int i = 0 ; i < nbWorkers ; ++i )
{
int wrkRank = idles[i];
Node::comm().send( wrkRank, 1, realSeeds[ i ] );
}
} else
{
int seed;
Node::comm().recv( masterRank, 1, seed );
std::cout << Node::comm().rank() << "- Seed: " << seed << std::endl;
eo::rng.reseed( seed );
}
store.init( algo.idles(), runs );
}
EOT& best_individual()
@ -303,95 +319,205 @@ class MultiStart : public MultiJob< MultiStartData< EOT, FitT > >
}
};
template<class EOT, class FitT>
struct DummyGetSeeds : public MultiStartStore<EOT,FitT>::GetSeeds
{
std::vector<int> operator()( int n )
{
return std::vector<int>();
}
};
template<class EOT, class FitT>
struct GetRandomSeeds : public MultiStartStore<EOT,FitT>::GetSeeds
{
std::vector<int> operator()( int n )
{
std::vector<int> ret;
for(int i = 0; i < n; ++i)
{
ret.push_back( eo::rng.rand() );
}
}
};
template<class EOT, class FitT>
struct ReinitMultiEA : public MultiStartStore<EOT,FitT>::ReinitJob
{
ReinitMultiEA( const eoPop<EOT>& pop, eoEvalFunc<EOT>& eval ) : _originalPop( pop ), _eval( eval )
{
// empty
}
void operator()( eoPop<EOT>& pop )
{
pop = _originalPop;
for(unsigned i = 0, size = pop.size(); i < size; ++i)
{
_eval( pop[i] );
}
}
private:
const eoPop<EOT>& _originalPop;
eoEvalFunc<EOT>& _eval;
};
template<class EOT, class FitT>
struct ResetAlgoEA : public MultiStartStore<EOT,FitT>::ResetAlgo
{
ResetAlgoEA( eoGenContinue<EOT> & continuator ) :
_continuator( continuator ),
_initial( continuator.totalGenerations() )
{
// empty
}
void operator()()
{
_continuator.totalGenerations( _initial );
}
private:
unsigned int _initial;
eoGenContinue<EOT> & _continuator;
};
template< class EOT >
struct eoInitAndEval : public eoInit<EOT>
{
eoInitAndEval( eoInit<EOT>& init, eoEvalFunc<EOT>& eval ) : _init( init ), _eval( eval )
{
// empty
}
void operator()( EOT & indi )
{
_init( indi );
_eval( indi );
}
private:
eoInit<EOT>& _init;
eoEvalFunc<EOT>& _eval;
};
int main(int argc, char **argv)
{
Node::init( argc, argv );
const unsigned int VEC_SIZE = 2;
const unsigned int POP_SIZE = 20;
const unsigned int NEIGHBORHOOD_SIZE= 5;
unsigned i;
// PARAMETRES
// all parameters are hard-coded!
const unsigned int SEED = 133742; // seed for random number generator
const unsigned int VEC_SIZE = 8; // Number of object variables in genotypes
const unsigned int POP_SIZE = 20; // Size of population
const unsigned int T_SIZE = 3; // size for tournament selection
const unsigned int MAX_GEN = 20; // Maximum number of generation before STOP
const float CROSS_RATE = 0.8; // Crossover rate
const double EPSILON = 0.01; // range for real uniform mutation
const float MUT_RATE = 0.5; // mutation rate
eo::rng.reseed(1);
// GENERAL
//////////////////////////
// Random seed
//////////////////////////
//reproducible random seed: if you don't change SEED above,
// you'll aways get the same result, NOT a random run
rng.reseed(SEED);
// the population:
eoPop<Particle> pop;
// EVAL
/////////////////////////////
// Fitness function
////////////////////////////
// Evaluation: from a plain C++ fn to an EvalFunc Object
eoEvalFuncPtr<Indi> eval( real_value );
// Evaluation
eoEvalFuncPtr<Particle, double, const Particle& > eval( real_value );
// INIT
////////////////////////////////
// Initilisation of population
////////////////////////////////
// position init
eoUniformGenerator < double >uGen (-3, 3);
eoInitFixedLength < Particle > random (VEC_SIZE, uGen);
// declare the population
eoPop<Indi> pop;
// fill it!
/*
for (unsigned int igeno=0; igeno<POP_SIZE; igeno++)
{
Indi v; // void individual, to be filled
for (unsigned ivar=0; ivar<VEC_SIZE; ivar++)
{
double r = 2*rng.uniform() - 1; // new value, random in [-1,1)
v.push_back(r); // append that random value to v
}
eval(v); // evaluate it
pop.push_back(v); // and put it in the population
}
*/
eoUniformGenerator< double > generator;
eoInitFixedLength< Indi > init( VEC_SIZE, generator );
// eoInitAndEval< Indi > init( real_init, eval, continuator );
pop = eoPop<Indi>( POP_SIZE, init );
// velocity init
eoUniformGenerator < double >sGen (-2, 2);
eoVelocityInitFixedLength < Particle > veloRandom (VEC_SIZE, sGen);
// ENGINE
/////////////////////////////////////
// selection and replacement
////////////////////////////////////
// SELECT
// The robust tournament selection
eoDetTournamentSelect<Indi> select(T_SIZE); // T_SIZE in [2,POP_SIZE]
// local best init
eoFirstIsBestInit < Particle > localInit;
// REPLACE
// eoSGA uses generational replacement by default
// so no replacement procedure has to be given
// perform position initialization
pop.append (POP_SIZE, random);
// OPERATORS
//////////////////////////////////////
// The variation operators
//////////////////////////////////////
// CROSSOVER
// offspring(i) is a linear combination of parent(i)
eoSegmentCrossover<Indi> xover;
// MUTATION
// offspring(i) uniformly chosen in [parent(i)-epsilon, parent(i)+epsilon]
eoUniformMutation<Indi> mutation(EPSILON);
// topology
eoLinearTopology<Particle> topology(NEIGHBORHOOD_SIZE);
// STOP
// CHECKPOINT
//////////////////////////////////////
// termination condition
/////////////////////////////////////
// stop after MAX_GEN generations
eoGenContinue<Indi> continuator(MAX_GEN); /** TODO FIXME FIXME BUG HERE!
Continuator thinks it's done! */
// the full initializer
eoInitializer <Particle> init(eval,veloRandom,localInit,topology,pop);
init();
// GENERATION
/////////////////////////////////////////
// the algorithm
////////////////////////////////////////
// standard Generational GA requires
// selection, evaluation, crossover and mutation, stopping criterion
// bounds
eoRealVectorBounds bnds(VEC_SIZE,-1.5,1.5);
// velocity
eoStandardVelocity <Particle> velocity (topology,1,1.6,2,bnds);
// flight
eoStandardFlight <Particle> flight;
// Terminators
eoGenContinue <Particle> genCont1 (50);
eoGenContinue <Particle> genCont2 (50);
// PS flight
eoEasyPSO<Particle> pso1(genCont1, eval, velocity, flight);
// eoEasyPSO<Particle> pso2(init,genCont2, eval, velocity, flight);
eoSGA<Indi> gga(select, xover, CROSS_RATE, mutation, MUT_RATE,
eval, continuator);
DynamicAssignmentAlgorithm assignmentAlgo;
MultiStartStore< Particle, ParticleFitness > store( pso1, DEFAULT_MASTER, pop );
MultiStartStore< Indi, IndiFitness > store(
gga,
DEFAULT_MASTER,
*new ReinitMultiEA< Indi, IndiFitness >( pop, eval ),
*new ResetAlgoEA< Indi, IndiFitness >( continuator ),
*new DummyGetSeeds< Indi, IndiFitness >());
MultiStart< Particle, ParticleFitness > msjob( assignmentAlgo, DEFAULT_MASTER, store, 5 );
MultiStart< Indi, IndiFitness > msjob( assignmentAlgo, DEFAULT_MASTER, store, 5 );
msjob.run();
if( msjob.isMaster() )
{
eo::mpi::EmptyJob tjob( assignmentAlgo, DEFAULT_MASTER );
std::cout << "Global best individual has fitness " << msjob.best_fitness() << std::endl;
}
// flight
/*
try
{
pso1(pop);
std::cout << "FINAL POPULATION AFTER PSO n°1:" << std::endl;
for (i = 0; i < pop.size(); ++i)
std::cout << "\t" << pop[i] << " " << pop[i].fitness() << std::endl;
pso2(pop);
std::cout << "FINAL POPULATION AFTER PSO n°2:" << std::endl;
for (i = 0; i < pop.size(); ++i)
std::cout << "\t" << pop[i] << " " << pop[i].fitness() << std::endl;
}
catch (std::exception& e)
{
std::cout << "exception: " << e.what() << std::endl;;
exit(EXIT_FAILURE);
}
*/
MultiStart< Indi, IndiFitness > msjob10( assignmentAlgo, DEFAULT_MASTER, store, 10 );
msjob10.run();
return 0;
}