(mu/rho +/, lambda) - Evolution Strategy (ES) algorithm. More...

#include <AlgEvolutionStrategy.h>

Inheritance diagram for EvolutionStrategy:

Public Member Functions
bool	advanceGeneration (StateP state, DemeP deme)
	Perform a single generation on a single deme. More...

bool	initialize (StateP state)
	Initialize the algorithm, read parameters from the system, do a sanity check. More...

void	registerParameters (StateP state)
	Register algorithm's parameters (if any). More...

Public Member Functions inherited from Algorithm
std::string	getName ()
	Get algorithm name. Each algorithm is uniquely identified with its name. More...

virtual bool	isParallel ()
	Is algorithm parallel (false by default for all algorithms not inheriting ParallelAlgorithm class). More...

bool	isImplicitParallel ()
	Is algorithm run in implicit parallel mode (in development, see tutorial). More...

virtual bool	advanceGeneration (StateP, DemeP)=0
	Perform a single generation on a single deme. More...

virtual bool	initialize (StateP)
	Initialize the algorithm, read parameters from the system, do a sanity check. More...

virtual void	registerParameters (StateP)
	Register algorithm's parameters (if any). More...

virtual bool	initializePopulation (StateP)
	Evaluate initial population (called by State::run before evolution starts). More...

virtual void	read ()

virtual bool	advanceGeneration (StateP state)
	Perform one generation of the algorithm on the whole population. More...

void	evaluate (IndividualP ind)
	Helper function: evaluate an individual. More...

uint	mutate (const std::vector< IndividualP > &pool)
	Helper function: send a vector of individuals to mutation. More...

uint	mutate (const IndividualP victim)
	Helper function: send a single individual to mutation. More...

void	replaceWith (IndividualP oldInd, IndividualP newInd)
	Helper function: replace an individual in current deme. More...

void	replaceWith (uint oldIndId, IndividualP newInd)
	Helper function: replace an individual at given position in current deme. More...

void	registerParallelParameters (StateP state)
	used only in parallel ECF More...

bool	initializeParallel (StateP state)
	used only in parallel ECF More...

virtual bool	advanceGeneration (StateP state)
	Parallel ECF: Perform one generation of the algorithm.

virtual void	bcastTermination (StateP state)
	Parallel ECF: broadcast termination to worker processes. More...

void	registerParallelParameters (StateP state)
	Parallel ECF: Register parameters for implicit parallel algorithm version.

bool	initializeParallel (StateP state)
	Parallel ECF: Initialize implicit parallel algorithm.

void	initializeImplicit (StateP state)
	Parallel ECF: Initialize implicit parallel mode. More...

bool	implicitParallelOperate (StateP state)
	Parallel ECF: Worker processes in implicit parallel algorithm. More...

void	evaluate (IndividualP ind)
	Parallel ECF: evaluate an individual. More...

void	implicitEvaluate (IndividualP ind)
	Parallel ECF: implicitly evaluate an individual (store for later evaluation in implicit parallel version). More...

uint	mutate (const IndividualP victim)
	Parallel ECF: immediately mutate an individual. More...

uint	mutate (const std::vector< IndividualP > &pool)
	Parallel ECF: mutate a vector of individuals. More...

uint	implicitMutate (IndividualP ind)
	Parallel ECF: implicitly mutate an individual (store for later mutation in implicit parallel version). More...

void	replaceWith (IndividualP oldInd, IndividualP newInd)
	Parallel ECF: replace an individual in a deme.

void	replaceWith (uint oldIndId, IndividualP newInd)
	Parallel ECF: replace an individual at given position in a deme. More...

bool	mate (IndividualP p1, IndividualP p2, IndividualP child)
	Helper function: crossover two individuals. More...

IndividualP	copy (IndividualP source)
	Helper function: make a copy of an individual. More...

bool	removeFrom (IndividualP victim, std::vector< IndividualP > &pool)
	Helper function: remove victim from pool of individual pointers. More...

bool	isMember (IndividualP single, std::vector< IndividualP > &pool)
	Helper function: check if individual is in the pool. More...

Static Protected Member Functions
static bool	compare (IndividualP first, IndividualP second)

Protected Attributes
uint	lambda_
	number of offspring More...

uint	rho_
	number of parents (1 or 2) More...

uint	mu_
	the size of the parent population More...

uint	subPopulations_
	how many parent populations are in a deme More...

bool	plusSelection_
	type of selection (plus or comma) More...

SelectionOperatorP	selBestOp_

SelectionOperatorP	selRandomOp_

Protected Attributes inherited from Algorithm
DemeP	activeDeme_

std::string	name_
	algorithm name More...

bool	bImplicitParallel_
	implicit parallel flag More...

std::vector< SelectionOperatorP >	selectionOp
	sel. operators used by algorithm More...

std::vector< IndividualP >	requests_

std::vector< IndividualP >	stored_
	individual vectors for implicit evaluation More...

std::vector< uint >	requestIds_
	individual indexes for implicit evaluation More...

std::vector< IndividualP >	demeCopy_

std::vector< IndividualP >	myJob_
	worker's individual vector More...

CommunicatorP	comm_

uint	jobSize_

uint	totalEvaluations_

uint	wastedEvaluations_

bool	bImplicitEvaluation_
	implicit evaluation flag More...

bool	bImplicitMutation_
	implicit mutation flag More...

bool	bSynchronized_
	is implicit paralelization synchronous More...

SelectionOperatorP	selBestOp

std::vector< IndividualP >	requestsMut_

std::vector< IndividualP >	receivedMut_
	individual vectors for implicit mutation More...

std::vector< uint >	requestMutIds_
	individual indexes for implicit mutation More...

IndividualP	currentBest_

std::vector< IndividualP >	storedInds_

std::vector< std::vector< IndividualP > >	sentInds_
	individuals sent for evaluation More...

std::vector< bool >	isConsistent_
	is individual (genotype-fitness pair) consistent More...

Additional Inherited Members
Public Attributes inherited from Algorithm
CrossoverP	crossover_
	sptr to container of crossover operators (set by the system) More...

MutationP	mutation_
	sptr to container of mutation operators (set by the system) More...

EvaluateOpP	evalOp_
	sptr to evaluation operator (set by the system) More...

StateP	state_

Protected Member Functions inherited from Algorithm
bool	registerParameter (StateP state, std::string name, voidP value, enum ECF::type T, std::string description="")
	Helper function: register a single parameter with the system. More...

voidP	getParameterValue (StateP state, std::string name)
	Helper function: get parameter value from the system. More...

void	storeIndividual (IndividualP)
	stores the individual (if it is consistent), resets consistency flag More...

void	storeGenotypes (std::vector< IndividualP > &)
	adds genotypes of individuals to 'sent' repository More...

void	setConsistency (IndividualP)
	denotes current individual as consistent More...

void	restoreIndividuals (std::vector< uint >)
	restores individuals whose fitness is received More...

void	restorePopulation ()
	restores inconsistent individuals to last consistent state More...

Detailed Description

(mu/rho +/, lambda) - Evolution Strategy (ES) algorithm.

This algorithm is genotype independent (it can be used with any Genotype).

The algorithm parameters are:

mu: the size of the parent population
lambda: number of offspring created in each iteration
rho: number of parents used to create an offspring; 1: mutation is used, 2: crossover
selection: may be "plus", which uses both parents and offspring or "comma", which uses just the offspring pool

Since the ECF defines "population.size" independently of the algorithm, this implementation actually runs multiple ES parent populations; the population.size must therefore be a multiple of parent pool size (mu). For instance, if mu = 5 and population.size = 30, then 6 ES subpopulations will be executed (no migration).

The algorithm flow:

 subpopulations = population.size / mu;
 single generation {
  repeat(for every subpopulation) {
   add mu individuals to the parent pool;
   create lambda offspring using random rho parents for each;
   if comma selection 
     create new parent pool with mu best from offspring pool;
   else
     create new parent pool with mu best from offspring and parents pool;
  }
 }

Definition at line 40 of file AlgEvolutionStrategy.h.

Constructor & Destructor Documentation

◆ EvolutionStrategy()

EvolutionStrategy::EvolutionStrategy ( )

Definition at line 9 of file AlgEvolutionStrategy.cpp.

Member Function Documentation

◆ advanceGeneration()

bool EvolutionStrategy::advanceGeneration	(	StateP	,
		DemeP
	)

virtual

Perform a single generation on a single deme.

Must be implemented by a specific algorithm class

Implements Algorithm.

Definition at line 87 of file AlgEvolutionStrategy.cpp.

◆ compare()

static bool EvolutionStrategy::compare	(	IndividualP	first,
		IndividualP	second
	)

inlinestaticprotected

Definition at line 56 of file AlgEvolutionStrategy.h.

◆ initialize()

bool EvolutionStrategy::initialize ( StateP )

virtual

Initialize the algorithm, read parameters from the system, do a sanity check.

Called by the system before the algorithm starts (Algorithm::advanceGeneration)

Reimplemented from Algorithm.

Definition at line 33 of file AlgEvolutionStrategy.cpp.

◆ registerParameters()

void EvolutionStrategy::registerParameters ( StateP )

virtual

Register algorithm's parameters (if any).

Called by the system before algorithm initialization (Algorithm::initialize)

Reimplemented from Algorithm.

Definition at line 20 of file AlgEvolutionStrategy.cpp.

Member Data Documentation

◆ lambda_

uint EvolutionStrategy::lambda_

protected

number of offspring

Definition at line 49 of file AlgEvolutionStrategy.h.

◆ mu_

uint EvolutionStrategy::mu_

protected

the size of the parent population

Definition at line 51 of file AlgEvolutionStrategy.h.

◆ plusSelection_

bool EvolutionStrategy::plusSelection_

protected

type of selection (plus or comma)

Definition at line 53 of file AlgEvolutionStrategy.h.

◆ rho_

uint EvolutionStrategy::rho_

protected

number of parents (1 or 2)

Definition at line 50 of file AlgEvolutionStrategy.h.

◆ selBestOp_

SelectionOperatorP EvolutionStrategy::selBestOp_

protected

Definition at line 54 of file AlgEvolutionStrategy.h.

◆ selRandomOp_

SelectionOperatorP EvolutionStrategy::selRandomOp_

protected

Definition at line 54 of file AlgEvolutionStrategy.h.

◆ subPopulations_

uint EvolutionStrategy::subPopulations_

protected

how many parent populations are in a deme

Definition at line 52 of file AlgEvolutionStrategy.h.

The documentation for this class was generated from the following files:

D:/Projekt/ECF_trunk/ECF/AlgEvolutionStrategy.h
D:/Projekt/ECF_trunk/ECF/AlgEvolutionStrategy.cpp

Public Member Functions

Static Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ EvolutionStrategy()

Member Function Documentation

◆ advanceGeneration()

◆ compare()

◆ initialize()

◆ registerParameters()

Member Data Documentation

◆ lambda_

◆ mu_

◆ plusSelection_

◆ rho_

◆ selBestOp_

◆ selRandomOp_

◆ subPopulations_