com.rapidminer.operator.performance
Class SimpleCriterion

java.lang.Object
  com.rapidminer.operator.AbstractIOObject
      com.rapidminer.operator.ResultObjectAdapter
          com.rapidminer.tools.math.Averagable
              com.rapidminer.operator.performance.PerformanceCriterion
                  com.rapidminer.operator.performance.MeasuredPerformance
                      com.rapidminer.operator.performance.SimpleCriterion

All Implemented Interfaces:

IOObject, ResultObject, Readable, Reportable, LoggingHandler, java.io.Serializable, java.lang.Cloneable, java.lang.Comparable<PerformanceCriterion>

Direct Known Subclasses:

AbsoluteError, LenientRelativeError, RelativeError, RootMeanSquaredError, SimpleClassificationError, SquaredError, StrictRelativeError

public abstract class SimpleCriterion
extends MeasuredPerformance

Simple criteria are those which error can be counted for each example and can be averaged by the number of examples. Since errors should be minimized, the fitness is calculated as -1 multiplied by the the error. Subclasses might also want to implement the method transform(double) which applies a transformation on the value sum divided by the number of counted examples. This is for example usefull in case of root_means_squared error. All subclasses can be used for both regression and classification problems. In case of classification the confidence value for the desired true label is used as prediction.

Author:

Ingo Mierswa, Simon Fischer

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class com.rapidminer.operator.AbstractIOObject

AbstractIOObject.InputStreamProvider

Constructor Summary

SimpleCriterion()

SimpleCriterion(SimpleCriterion sc)

Method Summary

void	buildSingleAverage(Averagable performance) This method should build the average of this and another averagable of the same type.
protected abstract double	countExample(double label, double predictedLabel) Subclasses must count the example and return the value to sum up.
void	countExample(Example example) Invokes countExample(double, double) and counts the deviation.
protected void	countExampleWithWeight(double deviation, double weight)
double	getExampleCount() Returns the number of data points which was used to determine the criterion value.
double	getFitness() Returns the fitness depending on the value.
double	getMaxFitness() Returns 0.0.
double	getMikroAverage() Returns the (current) value of the averagable (the average itself).
double	getMikroVariance() Returns the variance of the averagable.
void	startCounting(ExampleSet eset, boolean useExampleWeights) Initializes the criterion.
protected double	transform(double value) Simply returns the given value.

Methods inherited from class com.rapidminer.operator.performance.MeasuredPerformance

startCounting

Methods inherited from class com.rapidminer.operator.performance.PerformanceCriterion

compareTo, getDescription

Methods inherited from class com.rapidminer.tools.math.Averagable

buildAverage, clone, cloneAveragable, formatPercent, getAverage, getAverageCount, getExtension, getFileDescription, getMakroAverage, getMakroStandardDeviation, getMakroVariance, getMikroStandardDeviation, getName, getStandardDeviation, getVariance, isInTargetEncoding, setAverageCount, toString

Methods inherited from class com.rapidminer.operator.ResultObjectAdapter

addAction, getActions, getAnnotations, getResultIcon, log, log, logError, logNote, logWarning, toHTML, toResultString

Methods inherited from class com.rapidminer.operator.AbstractIOObject

appendOperatorToHistory, copy, getLog, getProcessingHistory, getSource, initWriting, read, read, read, read, setLoggingHandler, setSource, write

Methods inherited from class java.lang.Object

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

Methods inherited from interface com.rapidminer.operator.IOObject

appendOperatorToHistory, copy, getLog, getProcessingHistory, getSource, setLoggingHandler, setSource, write

Constructor Detail

SimpleCriterion

public SimpleCriterion()

SimpleCriterion

public SimpleCriterion(SimpleCriterion sc)

Method Detail

getExampleCount

public double getExampleCount()

Description copied from class: PerformanceCriterion

Returns the number of data points which was used to determine the criterion value. If the criterion does not use example weights (or no weight was given) then the returned value will be an integer. Otherwise, the returned value is the sum of all example weights.

Specified by:

getExampleCount in class PerformanceCriterion

countExample

public void countExample(Example example)

Invokes countExample(double, double) and counts the deviation. In case of a nominal label the confidence of the desired true label is used as prediction. For regression problems the usual predicted label is used.

Specified by:

countExample in class MeasuredPerformance

countExample

protected abstract double countExample(double label,
                                       double predictedLabel)

Subclasses must count the example and return the value to sum up.

transform

protected double transform(double value)

Simply returns the given value. Subclasses might apply a transformation on the error sum divided by the number of examples.

countExampleWithWeight

protected void countExampleWithWeight(double deviation,
                                      double weight)

getMikroAverage

public double getMikroAverage()

Description copied from class: Averagable

Returns the (current) value of the averagable (the average itself). If the method Averagable.buildSingleAverage(Averagable) was used, this method must return the micro average from both (or more) criteria. This is usually achieved by correctly implementing Averagable.buildSingleAverage(Averagable).

Specified by:

getMikroAverage in class Averagable

getMikroVariance

public double getMikroVariance()

Description copied from class: Averagable

Returns the variance of the averagable. The returned value must not be negative. If the averagable does not define a variance this method should return Double.NaN.

Specified by:

getMikroVariance in class Averagable

startCounting

public void startCounting(ExampleSet eset,
                          boolean useExampleWeights)
                   throws OperatorException

Description copied from class: MeasuredPerformance

Initializes the criterion. The default implementation does nothing.

Overrides:

startCounting in class MeasuredPerformance

Throws:

OperatorException

getFitness

public double getFitness()

Description copied from class: PerformanceCriterion

Returns the fitness depending on the value. The fitness values will be used for all optimization purposes (feature space transformations, parameter optimizations...) and must always be maximized. Hence, if your criterion is better the smaller the value is you should return something like (-1 * value) or (1 / value).

Subclasses should use Averagable.getAverage() instead of Averagable.getMikroAverage() in this method since usually the makro average (if available) should be optmized instead of the mikro average. The mikro average should only be used in the (rare) cases where no makro average is available but this is automatically done returned by Averagable.getAverage() in these cases.

Specified by:

getFitness in class PerformanceCriterion

getMaxFitness

public double getMaxFitness()

Returns 0.0.

Overrides:

getMaxFitness in class PerformanceCriterion

buildSingleAverage

public void buildSingleAverage(Averagable performance)

Description copied from class: Averagable

This method should build the average of this and another averagable of the same type. The next invocation of Averagable.getMikroAverage() should return the average of this and the given averagable. Hence, this method is used to build the actual micro average value of two criteria. Please refer to SimpleCriterion for a simple implementation example.

Specified by:

buildSingleAverage in class Averagable