com.epochx.core
Interface GPModel<TYPE>

Type Parameters:

TYPE - The return type of CandidatePrograms being evolved.

All Known Implementing Classes:

CubicRegression, Even4Parity, Even5Parity, Even7Parity, GPAbstractModel, Majority5, Majority9, Multiplexer11Bit, Multiplexer20Bit, Multiplexer37Bit, Multiplexer6Bit, QuarticRegression, SantaFeTrail

public interface GPModel<TYPE>

A GPModel defines all those parameters needed to control a run by GPRun. The first step - and for most problems the only step - to generate a GP evolved solution with EpochX is to provide a concrete implementation of this interface.

For most situations, users should look to extend the abstract GPAbstractModel.

See Also:

GPAbstractModel

Method Summary

boolean	acceptCrossover(CandidateProgram<TYPE>[] parents, CandidateProgram<TYPE>[] children) This method will be called during each crossover operation before the crossover is accepted, giving the model the opportunity to reject the operation, in which case the operation will be attempted again until it is accepted.
boolean	acceptMutation(CandidateProgram<TYPE> parent, CandidateProgram<TYPE> child) This method will be called during each mutation operation before the mutation is accepted, giving the model the opportunity to reject the operation, in which case the operation will be attempted again until it is accepted.
Crossover<TYPE>	getCrossover() Retrieves the implementation of Crossover to use to perform the genetic operation of crossover between 2 parents.
double	getCrossoverProbability() Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the crossover genetic operator, as opposed to mutation or reproduction.
CrossoverStatListener	getCrossoverStatListener() Get a listener which will be informed of statistics about crossovers.
double	getFitness(CandidateProgram<TYPE> program) Calculates and returns the fitness score of the given program.
java.util.List<FunctionNode<TYPE>>	getFunctions() Retrieves the set of function nodes.
GenerationStatListener	getGenerationStatListener() Get a listener which will be informed of statistics about generations.
Initialiser<TYPE>	getInitialiser() Retrieves the Initialiser which will generate the first generation population from which the evolution will proceed.
int	getInitialMaxDepth() Retrieves the maximum depth of CandidatePrograms allowed in the population after initialisation.
LifeCycleListener<TYPE>	getLifeCycleListener() Get a listener which will be informed of each stage of a GP run's life cycle, and given the facility to confirm or modify each step.
int	getMaxProgramDepth() Retrieves the maximum depth of CandidatePrograms allowed in the population after undergoing genetic operators.
double	getMutationProbability() Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the mutation genetic operator, as opposed to crossover or reproduction.
MutationStatListener	getMutationStatListener() Get a listener which will be informed of statistics about mutations.
Mutation<TYPE>	getMutator() Retrieves the implementation of Mutator to use to perform the genetic operation of mutation on a CandidateProgram.
int	getNoElites() Retrieves the number of elites that should be copied straight to the next population.
int	getNoGenerations() Retrieves the number of generations that each run should use before terminating, unless prior termination occurs due to one of the other termination criterion.
int	getNoRuns() Retrieves the number of runs that should be carried out using this model as the basis.
PoolSelector<TYPE>	getPoolSelector() Retrieves the selector to use to construct a breeding pool from which parents can be selected using the parent selector returned by getProgramSelector() to undergo the genetic operators.
int	getPoolSize() Retrieves the size of the breeding pool to be used for parent selection when performing the genetic operators.
int	getPopulationSize() Retrieves the number of CandidatePrograms that should make up each generation.
ProgramSelector<TYPE>	getProgramSelector() Retrieves the selector to use to pick parents from either a pre-selected breeding pool (selected by the PoolSelector returned by getPoolSelector()) or the previous population for use as input to the genetic operators.
double	getReproductionProbability() Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the reproduction genetic operator, as opposed to crossover or mutation.
RandomNumberGenerator	getRNG() Returns the RandomNumberGenerator instance that should be used for the generation of random numbers throughout execution.
RunStatListener	getRunStatListener() Get a listener which will be informed of statistics about runs.
java.util.List<Node<TYPE>>	getSyntax() Retrieves the full set of syntax, that is terminals AND function nodes.
java.util.List<TerminalNode<TYPE>>	getTerminals() Retrieves the set of terminal nodes.
double	getTerminationFitness() The fitness score at which a run should be stopped.

Method Detail

getInitialiser

Initialiser<TYPE> getInitialiser()

Retrieves the Initialiser which will generate the first generation population from which the evolution will proceed.

Returns:

the Initialiser to create the first population.

getCrossover

Crossover<TYPE> getCrossover()

Retrieves the implementation of Crossover to use to perform the genetic operation of crossover between 2 parents. The 2 parents to be crossed over will be selected using the parent selector returned by getProgramSelector().

Returns:

the implementation of Crossover that will perform the genetic operation of crossover.

See Also:

UniformPointCrossover, KozaCrossover

getMutator

Mutation<TYPE> getMutator()

Retrieves the implementation of Mutator to use to perform the genetic operation of mutation on a CandidateProgram. The individual to be mutated will be selected using the program selector returned by getProgramSelector().

Returns:

the implementation of Mutator that will perform the genetic operation of mutation.

getProgramSelector

ProgramSelector<TYPE> getProgramSelector()

Retrieves the selector to use to pick parents from either a pre-selected breeding pool (selected by the PoolSelector returned by getPoolSelector()) or the previous population for use as input to the genetic operators.

Returns:

the ProgramSelector which should be used to pick parents for input to the genetic operators.

See Also:

TournamentSelector

getPoolSelector

PoolSelector<TYPE> getPoolSelector()

Retrieves the selector to use to construct a breeding pool from which parents can be selected using the parent selector returned by getProgramSelector() to undergo the genetic operators.

Returns:

a PoolSelector which can be used to construct a breeding pool, or null if a breeding pool shouldn't be used and instead parents should be picked straight from the previous population.

See Also:

TournamentSelector

getTerminals

java.util.List<TerminalNode<TYPE>> getTerminals()

Retrieves the set of terminal nodes.

Returns:

the terminal nodes to be used during evolution.

getFunctions

java.util.List<FunctionNode<TYPE>> getFunctions()

Retrieves the set of function nodes.

Returns:

the function nodes to be used during evolution.

getSyntax

java.util.List<Node<TYPE>> getSyntax()

Retrieves the full set of syntax, that is terminals AND function nodes. This is usually a combination of the calls from getTerminals() and getFunctions().

Returns:

the full syntax for use in building node trees.

getRNG

RandomNumberGenerator getRNG()

Returns the RandomNumberGenerator instance that should be used for the generation of random numbers throughout execution.

Returns:

the RandomNumberGenerator to use for generating randomness.

getNoRuns

int getNoRuns()

Retrieves the number of runs that should be carried out using this model as the basis. Each call to GPRun.run() will be with the same model so this is useful when multiple runs are necessary with the same control parameters for research purposes or otherwise in order to attain reliable results drawn from mean averages.

Returns:

the number of times this model should be used to control GP runs.

getNoGenerations

int getNoGenerations()

Retrieves the number of generations that each run should use before terminating, unless prior termination occurs due to one of the other termination criterion.

Returns:

the number of generations that should be evolved in each run.

getPopulationSize

int getPopulationSize()

Retrieves the number of CandidatePrograms that should make up each generation. The population at the start and end of each generation should always equal exactly this number.

Returns:

the number of CandidatePrograms per generation.

getPoolSize

int getPoolSize()

Retrieves the size of the breeding pool to be used for parent selection when performing the genetic operators. If the pool size is equal to or less than zero, or if getPoolSelector() returns null, then no pool will be used and parent selection will take place directly from the previous population.

Returns:

the size of the mating pool to build with the PoolSelector returned by getPoolSelector() which will be used for parent selection.

getNoElites

int getNoElites()

Retrieves the number of elites that should be copied straight to the next population. This number is distinct from the reproduction operator. Elites are generally picked as the very best programs in a generation, thus a number of elites of 1 or more will always retain the best program found so far, through to the last generation.

Returns:

the number of elites that should be copied through from one generation to the next.

getInitialMaxDepth

int getInitialMaxDepth()

Retrieves the maximum depth of CandidatePrograms allowed in the population after initialisation. The exact way in which the implementation ensures this depth is kept to may vary.

Returns:

the maximum depth of CandidatePrograms to be allowed in the population after initialisation.

getMaxProgramDepth

int getMaxProgramDepth()

Retrieves the maximum depth of CandidatePrograms allowed in the population after undergoing genetic operators. The exact way in which CandidatePrograms deeper than this limit are dealt with may vary, but they will not be allowed to remain into the next generation unaltered.

Returns:

the maximum depth of CandidatePrograms to be allowed in the population after genetic operators.

getCrossoverProbability

double getCrossoverProbability()

Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the crossover genetic operator, as opposed to mutation or reproduction.

Within one generation approximately Pc proportion of the CandidatePrograms will have been created through crossover, Pm through mutation and Pr through reproduction. Where Pc, Pm, Pr are the values returned by getCrossoverProbability(), getMutationProbability() and getReproductionProbability() respectively. The sum of the calls to getCrossoverProbability(), getReproductionProbability() and getMutationProbability() should total 1.0.

Returns:

the probability of choosing the crossover genetic operator at each iteration when constructing the next population.

getMutationProbability

double getMutationProbability()

Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the mutation genetic operator, as opposed to crossover or reproduction.

Within one generation approximately Pc proportion of the CandidatePrograms will have been created through crossover, Pm through mutation and Pr through reproduction. Where Pc, Pm, Pr are the values returned by getCrossoverProbability(), getMutationProbability() and getReproductionProbability() respectively. The sum of the calls to getCrossoverProbability(), getReproductionProbability() and getMutationProbability() should total 1.0.

Returns:

the probability of choosing the mutation genetic operator at each iteration when constructing the next population.

getReproductionProbability

double getReproductionProbability()

Retrieves a numerical value between 0.0 and 1.0 which represents the probability of selecting the reproduction genetic operator, as opposed to crossover or mutation.

Within one generation approximately Pc proportion of the CandidatePrograms will have been created through crossover, Pm through mutation and Pr through reproduction. Where Pc, Pm, Pr are the values returned by getCrossoverProbability(), getMutationProbability() and getReproductionProbability() respectively. The sum of the calls to getCrossoverProbability(), getReproductionProbability() and getMutationProbability() should total 1.0.

Returns:

the probability of choosing the reproduction genetic operator at each iteration when constructing the next population.

getFitness

double getFitness(CandidateProgram<TYPE> program)

Calculates and returns the fitness score of the given program. The score returned by this method provides the underlying way in which Candidate Programs are selected. The GP system will attempt to improve the value returned by this method over the generations. A fitness value of 0.0 is better than a fitness value of 1.0.

There are many ways in which a fitness score can be calculated, for example mean squared error, standard deviation or a simple count of how many known inputs the given program provides incorrect (known) outputs. For more information, the new user should read some of the genetic programming literature.

Parameters:

program - the CandidateProgram to evaluate and calculate a score for.

Returns:

a double representing the quality of the program where a small value is considered better than a larger value.

getTerminationFitness

double getTerminationFitness()

The fitness score at which a run should be stopped. Returning a negative value will result in no termination based upon fitness.

Returns:

the fitness score at which a run should be terminated. A fitness of this or less will result in termination.

acceptCrossover

boolean acceptCrossover(CandidateProgram<TYPE>[] parents,
                        CandidateProgram<TYPE>[] children)

This method will be called during each crossover operation before the crossover is accepted, giving the model the opportunity to reject the operation, in which case the operation will be attempted again until it is accepted. The number of times crossovers are rejected is retrievable using the REVERTED_CROSSOVERS stats field.

Parameters:

parents - The programs which have been crossed over to create the given children.

children - The children which are the result of the crossover operation having been performed on the given parents.

Returns:

True if the crossover operation should proceed, false if it is rejected and should be retried with new parents.

acceptMutation

boolean acceptMutation(CandidateProgram<TYPE> parent,
                       CandidateProgram<TYPE> child)

This method will be called during each mutation operation before the mutation is accepted, giving the model the opportunity to reject the operation, in which case the operation will be attempted again until it is accepted. The number of times mutations are rejected is retrievable using the REVERTED_MUTATIONS stats field.

Parameters:

parent - The program before the mutation operation.

child - The program after the mutation operation has been carried out.

Returns:

True if the mutation operation should proceed, false if it is rejected and should be retried with a new parent.

getRunStatListener

RunStatListener getRunStatListener()

Get a listener which will be informed of statistics about runs. The listener will be queried for what fields are of interest, with those statistics passed to the runStats method in the same order at the end of each run.

Returns:

A RunStatListener to handle run statistics.

getGenerationStatListener

GenerationStatListener getGenerationStatListener()

Get a listener which will be informed of statistics about generations. The listener will be queried for what fields are of interest, with those statistics passed to the generationStats method in the same order after each generation.

Returns:

A GenerationStatListener to handle generation statistics.

getCrossoverStatListener

CrossoverStatListener getCrossoverStatListener()

Get a listener which will be informed of statistics about crossovers. The listener will be queried for what fields are of interest, with those statistics passed to the crossoverStats method in the same order after each crossover operation.

Returns:

A CrossoverStatListener to handle crossover statistics.

getMutationStatListener

MutationStatListener getMutationStatListener()

Get a listener which will be informed of statistics about mutations. The listener will be queried for what fields are of interest, with those statistics passed to the mutationStats method in the same order after each mutation operation.

Returns:

A MutationStatListener to handle mutation statistics.

getLifeCycleListener

LifeCycleListener<TYPE> getLifeCycleListener()

Get a listener which will be informed of each stage of a GP run's life cycle, and given the facility to confirm or modify each step.

Returns:

A LifeCycleListener to confirm or modify each stage of the run's life cycle.