edu.udo.cs.yale.tools.math.optimization.ec.es
Class ESOptimization

java.lang.Object
  edu.udo.cs.yale.tools.math.optimization.ec.es.ESOptimization

All Implemented Interfaces:

Optimization

Direct Known Subclasses:

ESParameterOptimization, EvoOptimization

public abstract class ESOptimization
extends java.lang.Object
implements Optimization

Evolutionary Strategy approach for SVM optimization. Currently only classification problems are supported.

Version:

$Id: ESOptimization.java,v 1.5 2006/09/06 10:37:35 ingomierswa Exp $

Author:

Ingo Mierswa

Field Summary

private int	actualEvalCounter This field counts the number of actually calculated evaluations (unchanged individuals do not have to be re-evaluated).
static int	BOLTZMANN_SELECTION Indicates a Boltzmann selection scheme.
static int	CUT_SELECTION Indicates a cut selection scheme.
static int	GAUSSIAN_MUTATION Indicates a gaussian mutation.
private int	generationsWithoutImprovement The maximum numbers of generations without improvement.
private int	individualSize The dimension of each individual.
static int	INIT_TYPE_MAX Indicates that the start population should be initialized with the maximum value.
static int	INIT_TYPE_MIN Indicates that the start population should be initialized with the minimum value.
static int	INIT_TYPE_RANDOM Indicates that the start population should be randomly initialized.
private int	initType The type of start population initialization.
private double[]	max This parameter indicates the maximum value for all genes.
private int	maxGenerations The maximum number of generations.
private double[]	min This parameter indicates the minimum value for all genes.
private PopulationOperator	mutation The mutation operator.
static java.lang.String[]	MUTATION_TYPES The names of the mutation types.
private int	mutationType The type of the mutation.
static int	NO_MUTATION Indicates no mutation.
static int	NON_DOMINATED_SORTING_SELECTION Indicates a multi-objective selection scheme (NSGA II).
private java.util.Collection<PopulationOperator>	popOps Population operators.
private Population	population The current population.
static java.lang.String[]	POPULATION_INIT_TYPES The names of the initialization types.
private int	populationSize The number of individuals.
private RandomGenerator	random The random number generator.
static int	RANK_SELECTION Indicates a rank based selection scheme.
static int	ROULETTE_WHEEL Indicates a roulette wheel selection scheme.
static java.lang.String[]	SELECTION_TYPES The names of all available selection schemes.
private boolean	showPlot Indicates if a convergence plot should be drawn.
static int	SPARSITY_MUTATION Indicates a hybrid between switching mutation and Gaussian mutation.
static int	STOCHASTIC_UNIVERSAL Indicates a stochastic universal sampling selection scheme.
static int	SWITCHING_MUTATION Indicates a switching mutation.
private int	totalEvalCounter This field counts the total number of evaluations during optimization.
static int	TOURNAMENT_SELECTION Indicates a tournament selection scheme.
static int	UNIFORM_SELECTION Indicates a uniform sampling selection scheme.
static int	VALUE_TYPE_DOUBLE This constant indicated that a corresponding dimension is actually built from double values.
static int	VALUE_TYPE_INT This constant indicates that the dimension is actually built from integers instead of doubles.
private int[]	valueTypes The value types, either DOUBLE (default) or INT.

Constructor Summary

ESOptimization(double minValue, double maxValue, int populationSize, int individualSize, int initType, int maxGenerations, int generationsWithoutImprovement, int selectionType, double tournamentFraction, boolean keepBest, int mutationType, double crossoverProb, boolean showPlot, RandomGenerator random)
Creates a new evolutionary SVM optimization.

Method Summary

private Population	createMaxStartPopulation() Randomly creates the initial population.
private Population	createMinStartPopulation() Randomly creates the initial population.
private Population	createRandomStartPopulation() Randomly creates the initial population.
private void	evaluate(Population population) Evaluates the individuals of the given population.
abstract PerformanceVector	evaluateIndividual(Individual individual) Subclasses must implement this method to calculate the fitness of the given individual.
double	getBestFitnessEver() Returns the best fitness ever.
double	getBestFitnessInGeneration() Returns the best fitness in the current generation.
PerformanceVector	getBestPerformanceEver() Returns the best performance vector ever.
double[]	getBestValuesEver() Returns the best values ever.
int	getGeneration() Returns the current generation.
double	getMax(int index)
double	getMin(int index)
int	getValueType(int index)
void	nextIteration() This method is invoked after each evaluation.
void	optimize() Starts the optimization.
private void	recalculateSigma()
void	setMax(int index, double v)
void	setMin(int index, double v)
void	setValueType(int index, int type)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

VALUE_TYPE_DOUBLE

public static final int VALUE_TYPE_DOUBLE

This constant indicated that a corresponding dimension is actually built from double values.

See Also:

Constant Field Values

VALUE_TYPE_INT

public static final int VALUE_TYPE_INT

This constant indicates that the dimension is actually built from integers instead of doubles.

See Also:

Constant Field Values

SELECTION_TYPES

public static final java.lang.String[] SELECTION_TYPES

The names of all available selection schemes.

UNIFORM_SELECTION

public static final int UNIFORM_SELECTION

Indicates a uniform sampling selection scheme.

See Also:

Constant Field Values

CUT_SELECTION

public static final int CUT_SELECTION

Indicates a cut selection scheme.

See Also:

Constant Field Values

ROULETTE_WHEEL

public static final int ROULETTE_WHEEL

Indicates a roulette wheel selection scheme.

See Also:

Constant Field Values

STOCHASTIC_UNIVERSAL

public static final int STOCHASTIC_UNIVERSAL

Indicates a stochastic universal sampling selection scheme.

See Also:

Constant Field Values

BOLTZMANN_SELECTION

public static final int BOLTZMANN_SELECTION

Indicates a Boltzmann selection scheme.

See Also:

Constant Field Values

RANK_SELECTION

public static final int RANK_SELECTION

Indicates a rank based selection scheme.

See Also:

Constant Field Values

TOURNAMENT_SELECTION

public static final int TOURNAMENT_SELECTION

Indicates a tournament selection scheme.

See Also:

Constant Field Values

NON_DOMINATED_SORTING_SELECTION

public static final int NON_DOMINATED_SORTING_SELECTION

Indicates a multi-objective selection scheme (NSGA II).

See Also:

Constant Field Values

MUTATION_TYPES

public static final java.lang.String[] MUTATION_TYPES

The names of the mutation types.

NO_MUTATION

public static final int NO_MUTATION

Indicates no mutation.

See Also:

Constant Field Values

GAUSSIAN_MUTATION

public static final int GAUSSIAN_MUTATION

Indicates a gaussian mutation.

See Also:

Constant Field Values

SWITCHING_MUTATION

public static final int SWITCHING_MUTATION

Indicates a switching mutation.

See Also:

Constant Field Values

SPARSITY_MUTATION

public static final int SPARSITY_MUTATION

Indicates a hybrid between switching mutation and Gaussian mutation.

See Also:

Constant Field Values

POPULATION_INIT_TYPES

public static final java.lang.String[] POPULATION_INIT_TYPES

The names of the initialization types.

INIT_TYPE_RANDOM

public static final int INIT_TYPE_RANDOM

Indicates that the start population should be randomly initialized.

See Also:

Constant Field Values

INIT_TYPE_MIN

public static final int INIT_TYPE_MIN

Indicates that the start population should be initialized with the minimum value.

See Also:

Constant Field Values

INIT_TYPE_MAX

public static final int INIT_TYPE_MAX

Indicates that the start population should be initialized with the maximum value.

See Also:

Constant Field Values

min

private double[] min

This parameter indicates the minimum value for all genes.

max

private double[] max

This parameter indicates the maximum value for all genes.

valueTypes

private int[] valueTypes

The value types, either DOUBLE (default) or INT.

populationSize

private int populationSize

The number of individuals.

individualSize

private int individualSize

The dimension of each individual.

maxGenerations

private int maxGenerations

The maximum number of generations.

generationsWithoutImprovement

private int generationsWithoutImprovement

The maximum numbers of generations without improvement.

initType

private int initType

The type of start population initialization.

mutationType

private int mutationType

The type of the mutation.

mutation

private PopulationOperator mutation

The mutation operator.

population

private Population population

The current population.

popOps

private java.util.Collection<PopulationOperator> popOps

Population operators.

showPlot

private boolean showPlot

Indicates if a convergence plot should be drawn.

totalEvalCounter

private int totalEvalCounter

This field counts the total number of evaluations during optimization.

actualEvalCounter

private int actualEvalCounter

This field counts the number of actually calculated evaluations (unchanged individuals do not have to be re-evaluated).

random

private RandomGenerator random

The random number generator.

Constructor Detail

ESOptimization

public ESOptimization(double minValue,
                      double maxValue,
                      int populationSize,
                      int individualSize,
                      int initType,
                      int maxGenerations,
                      int generationsWithoutImprovement,
                      int selectionType,
                      double tournamentFraction,
                      boolean keepBest,
                      int mutationType,
                      double crossoverProb,
                      boolean showPlot,
                      RandomGenerator random)

Creates a new evolutionary SVM optimization.

Method Detail

evaluateIndividual

public abstract PerformanceVector evaluateIndividual(Individual individual)
                                              throws OperatorException

Subclasses must implement this method to calculate the fitness of the given individual. Please note that null might be returned for non-valid individuals. The fitness will be maximized.

Throws:

OperatorException

nextIteration

public void nextIteration()
                   throws OperatorException

This method is invoked after each evaluation. The default implementation does nothing but subclasses might implement this method to support online plotting or logging.

Throws:

OperatorException

getMin

public double getMin(int index)

getMax

public double getMax(int index)

setMin

public void setMin(int index,
                   double v)

setMax

public void setMax(int index,
                   double v)

recalculateSigma

private void recalculateSigma()

getValueType

public int getValueType(int index)

setValueType

public void setValueType(int index,
                         int type)

optimize

public void optimize()
              throws OperatorException

Starts the optimization.

Specified by:

optimize in interface Optimization

Throws:

OperatorException

evaluate

private void evaluate(Population population)
               throws OperatorException

Evaluates the individuals of the given population.

Throws:

OperatorException

getGeneration

public int getGeneration()

Returns the current generation.

Specified by:

getGeneration in interface Optimization

getBestFitnessInGeneration

public double getBestFitnessInGeneration()

Returns the best fitness in the current generation.

Specified by:

getBestFitnessInGeneration in interface Optimization

getBestFitnessEver

public double getBestFitnessEver()

Returns the best fitness ever.

Specified by:

getBestFitnessEver in interface Optimization

getBestPerformanceEver

public PerformanceVector getBestPerformanceEver()

Returns the best performance vector ever.

Specified by:

getBestPerformanceEver in interface Optimization

getBestValuesEver

public double[] getBestValuesEver()

Returns the best values ever. Use this method after optimization to get the best result.

Specified by:

getBestValuesEver in interface Optimization

createRandomStartPopulation

private Population createRandomStartPopulation()

Randomly creates the initial population.

createMinStartPopulation

private Population createMinStartPopulation()

Randomly creates the initial population.

createMaxStartPopulation

private Population createMaxStartPopulation()

Randomly creates the initial population.