com.rapidminer.operator.meta
Class EvolutionaryParameterOptimizationOperator

java.lang.Object
  com.rapidminer.operator.Operator
      com.rapidminer.operator.OperatorChain
          com.rapidminer.operator.meta.ParameterIteratingOperatorChain
              com.rapidminer.operator.meta.ParameterOptimizationOperator
                  com.rapidminer.operator.meta.EvolutionaryParameterOptimizationOperator

All Implemented Interfaces:

ConfigurationListener, PreviewListener, ParameterHandler, LoggingHandler

public class EvolutionaryParameterOptimizationOperator
extends ParameterOptimizationOperator

This operator finds the optimal values for a set of parameters using an evolutionary strategies approach which is often more appropriate than a grid search or a greedy search like the quadratic programming approach and leads to better results. The parameter parameters is a list of key value pairs where the keys are of the form operator_name.parameter_name and the value for each parameter must be a semicolon separated pair of a minimum and a maximum value in squared parantheses, e.g. [10;100] for a range of 10 until 100.
The operator returns an optimal ParameterSet which can as well be written to a file with a ParameterSetWriter. This parameter set can be read in another process using a ParameterSetLoader.
The file format of the parameter set file is straightforward and can easily be generated by external applications. Each line is of the form

operator_name.parameter_name = value
Please refer to section Advanced Processes/Parameter and performance analysis for an example application.

Author:

Ingo Mierswa, Tobias Malbrecht

Field Summary

static java.lang.String	PARAMETER_CROSSOVER_PROB The parameter name for "The probability for crossover.
static java.lang.String	PARAMETER_GENERATIONS_WITHOUT_IMPROVAL The parameter name for "Stop after this number of generations without improvement (-1: optimize until max_iterations).
static java.lang.String	PARAMETER_KEEP_BEST The parameter name for "Indicates if the best individual should survive (elititst selection).
static java.lang.String	PARAMETER_LOCAL_RANDOM_SEED The parameter name for "Use the given random seed instead of global random numbers (-1: use global).
static java.lang.String	PARAMETER_MAX_GENERATIONS The parameter name for "Stop after this many evaluations"
static java.lang.String	PARAMETER_MUTATION_TYPE The parameter name for "The type of the mutation operator.
static java.lang.String	PARAMETER_POPULATION_SIZE The parameter name for "The population size (-1: number of examples)"
static java.lang.String	PARAMETER_SELECTION_TYPE The parameter name for "The type of the selection operator.
static java.lang.String	PARAMETER_SHOW_CONVERGENCE_PLOT The parameter name for "Indicates if a dialog with a convergence plot should be drawn.
static java.lang.String	PARAMETER_TOURNAMENT_FRACTION The parameter name for "The fraction of the population used for tournament selection.

Fields inherited from class com.rapidminer.operator.meta.ParameterIteratingOperatorChain

PARAMETER_PARAMETERS, PARAMETER_VALUES, VALUE_MODE_CONTINUOUS, VALUE_MODE_DISCRETE

Constructor Summary

EvolutionaryParameterOptimizationOperator(OperatorDescription description)

Method Summary

IOObject[]	apply() Applies all inner operators.
protected ESOptimization	createOptimizer(RandomGenerator random) This method creates a apropriate optimizer
double	getCurrentBestPerformance()
int	getNumberOfOptimizationParameters()
ESOptimization	getOptimization()
IOContainer	getOptimizationInput()
Operator[]	getOptimizationOperators()
java.lang.String[]	getOptimizationParameters()
OptimizationValueType[]	getOptimizationValueTypes()
java.util.List<ParameterType>	getParameterTypes() Returns a list of ParameterTypes describing the parameters of this operator.
int	getParameterValueMode() Has to return one of the predefined modes which indicate whether the operator takes discrete values or intervals as basis for optimization.
PerformanceVector	setParametersAndEvaluate(Individual individual)

Methods inherited from class com.rapidminer.operator.meta.ParameterOptimizationOperator

getInnerOperatorCondition, getInputClasses, getOutputClasses, getPerformance, getPerformance

Methods inherited from class com.rapidminer.operator.meta.ParameterIteratingOperatorChain

getMaxNumberOfInnerOperators, getMinNumberOfInnerOperators, parseParameterValues, shouldAddNonConsumedInput

Methods inherited from class com.rapidminer.operator.OperatorChain

addAddListener, addOperator, addOperator, checkDeprecations, checkIO, checkNumberOfInnerOperators, checkProperties, clearErrorList, cloneOperator, createExperimentTree, createProcessTree, getAllInnerOperators, getIndexOfOperator, getInnerOperatorForName, getInnerOperatorsXML, getNumberOfAllOperators, getNumberOfOperators, getOperator, getOperatorFromAll, getOperators, performAdditionalChecks, processFinished, processStarts, registerOperator, removeAddListener, removeOperator, shouldReturnInnerOutput, unregisterOperator

Methods inherited from class com.rapidminer.operator.Operator

addError, addValue, addWarning, apply, checkForStop, createExperimentTree, createFromXML, createMarkedExperimentTree, createMarkedProcessTree, createProcessTree, getAddOnlyAdditionalOutput, getApplyCount, getDeliveredOutputClasses, getDeprecationInfo, getDesiredInputClasses, getEncoding, getErrorList, getExperiment, getInput, getInput, getInput, getInputDescription, getIOContainerForInApplyLoopBreakpoint, getIODescription, getLog, getName, getOperatorClassName, getOperatorDescription, getParameter, getParameterAsBoolean, getParameterAsColor, getParameterAsDouble, getParameterAsFile, getParameterAsFile, getParameterAsInputStream, getParameterAsInt, getParameterAsMatrix, getParameterAsString, getParameterList, getParameters, getParameterType, getParent, getProcess, getStartTime, getStatus, getUserDescription, getValue, getValues, getXML, hasBreakpoint, hasBreakpoint, hasInput, inApplyLoop, isDebugMode, isEnabled, isExpanded, isParallel, isParameterSet, log, logError, logNote, logWarning, register, remove, rename, resume, setApplyCount, setBreakpoint, setEnabled, setExpanded, setInput, setListParameter, setOperatorParameters, setParameter, setParameters, setParent, setUserDescription, toString, writeXML

Methods inherited from class java.lang.Object

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

Field Detail

PARAMETER_MAX_GENERATIONS

public static final java.lang.String PARAMETER_MAX_GENERATIONS

The parameter name for "Stop after this many evaluations"

See Also:

Constant Field Values

PARAMETER_GENERATIONS_WITHOUT_IMPROVAL

public static final java.lang.String PARAMETER_GENERATIONS_WITHOUT_IMPROVAL

The parameter name for "Stop after this number of generations without improvement (-1: optimize until max_iterations)."

See Also:

Constant Field Values

PARAMETER_POPULATION_SIZE

public static final java.lang.String PARAMETER_POPULATION_SIZE

The parameter name for "The population size (-1: number of examples)"

See Also:

Constant Field Values

PARAMETER_TOURNAMENT_FRACTION

public static final java.lang.String PARAMETER_TOURNAMENT_FRACTION

The parameter name for "The fraction of the population used for tournament selection."

See Also:

Constant Field Values

PARAMETER_KEEP_BEST

public static final java.lang.String PARAMETER_KEEP_BEST

The parameter name for "Indicates if the best individual should survive (elititst selection)."

See Also:

Constant Field Values

PARAMETER_MUTATION_TYPE

public static final java.lang.String PARAMETER_MUTATION_TYPE

The parameter name for "The type of the mutation operator."

See Also:

Constant Field Values

PARAMETER_SELECTION_TYPE

public static final java.lang.String PARAMETER_SELECTION_TYPE

The parameter name for "The type of the selection operator."

See Also:

Constant Field Values

PARAMETER_CROSSOVER_PROB

public static final java.lang.String PARAMETER_CROSSOVER_PROB

The parameter name for "The probability for crossover."

See Also:

Constant Field Values

PARAMETER_LOCAL_RANDOM_SEED

public static final java.lang.String PARAMETER_LOCAL_RANDOM_SEED

The parameter name for "Use the given random seed instead of global random numbers (-1: use global)."

See Also:

Constant Field Values

PARAMETER_SHOW_CONVERGENCE_PLOT

public static final java.lang.String PARAMETER_SHOW_CONVERGENCE_PLOT

The parameter name for "Indicates if a dialog with a convergence plot should be drawn."

See Also:

Constant Field Values

Constructor Detail

EvolutionaryParameterOptimizationOperator

public EvolutionaryParameterOptimizationOperator(OperatorDescription description)

Method Detail

getOptimizationInput

public IOContainer getOptimizationInput()

getOptimizationOperators

public Operator[] getOptimizationOperators()

getOptimizationParameters

public java.lang.String[] getOptimizationParameters()

getOptimizationValueTypes

public OptimizationValueType[] getOptimizationValueTypes()

getParameterValueMode

public int getParameterValueMode()

Description copied from class: ParameterIteratingOperatorChain

Has to return one of the predefined modes which indicate whether the operator takes discrete values or intervals as basis for optimization. The first option is to be taken for all strategies that iterate over the given parameters. The latter option is to be taken for strategies such as an evolutionary one in which allowed ranges of parameters have to be specified.

Specified by:

getParameterValueMode in class ParameterIteratingOperatorChain

getCurrentBestPerformance

public double getCurrentBestPerformance()

Specified by:

getCurrentBestPerformance in class ParameterOptimizationOperator

apply

public IOObject[] apply()
                 throws OperatorException

Description copied from class: OperatorChain

Applies all inner operators. The input to this operator becomes the input of the first inner operator. The latter's output is passed to the second inner operator and so on. Note to subclassers: If subclasses (for example wrappers) want to make use of this method remember to call exactly this method (super.apply()) and do not call super.apply(IOContainer) erroneously which will result in an infinite loop.

Overrides:

apply in class OperatorChain

Returns:

the last inner operator's output or the input itself if the chain is empty.

Throws:

OperatorException

createOptimizer

protected ESOptimization createOptimizer(RandomGenerator random)
                                  throws UndefinedParameterError

This method creates a apropriate optimizer

Throws:

UndefinedParameterError

getOptimization

public ESOptimization getOptimization()

setParametersAndEvaluate

public PerformanceVector setParametersAndEvaluate(Individual individual)
                                           throws OperatorException

Throws:

OperatorException

getParameterTypes

public java.util.List<ParameterType> getParameterTypes()

Description copied from class: Operator

Returns a list of ParameterTypes describing the parameters of this operator. The default implementation returns an empty list if no input objects can be retained and special parameters for those input objects which can be prevented from being consumed.

Specified by:

getParameterTypes in interface ParameterHandler

Overrides:

getParameterTypes in class ParameterIteratingOperatorChain

getNumberOfOptimizationParameters

public int getNumberOfOptimizationParameters()