weka.distributed

Class WekaClassifierReduceTask

java.lang.Object

weka.distributed.WekaClassifierReduceTask

All Implemented Interfaces:

java.io.Serializable

public class WekaClassifierReduceTask
extends java.lang.Object
implements java.io.Serializable

Reduce task for aggregating classifiers into one final model, if they all directly implement Aggregateable, or into a voted ensemble otherwise

Version:

$Revision: 10652 $

Author:

Mark Hall (mhall{[at]}pentaho{[dot]}com)

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
WekaClassifierReduceTask()

Method Summary

Modifier and Type	Method and Description
weka.classifiers.Classifier	aggregate(java.util.List<weka.classifiers.Classifier> classifiers) Aggregate the supplied list of classifiers
weka.classifiers.Classifier	aggregate(java.util.List<weka.classifiers.Classifier> classifiers, java.util.List<java.lang.Integer> numTrainingInstancesPerClassifier, boolean forceVote) Aggregated the supplied list of classifiers.
java.util.List<java.lang.Integer>	getDiscarded() Get list of indices of the classifiers that were discarded (if any)
double	getMinTrainingFraction() Get the minimum training fraction by which a classifier is discarded.
void	setMinTrainingFraction(double m) Set the minimum training fraction by which a classifier is discarded.

Methods inherited from class java.lang.Object

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

Constructor Detail

WekaClassifierReduceTask

public WekaClassifierReduceTask()

Method Detail

aggregate

public weka.classifiers.Classifier aggregate(java.util.List<weka.classifiers.Classifier> classifiers)
                                      throws DistributedWekaException

Aggregate the supplied list of classifiers

Parameters:

classifiers - the classifiers to aggregate

Returns:

the final aggregated classifier

Throws:

DistributedWekaException - if a problem occurs

aggregate

public weka.classifiers.Classifier aggregate(java.util.List<weka.classifiers.Classifier> classifiers,
                                             java.util.List<java.lang.Integer> numTrainingInstancesPerClassifier,
                                             boolean forceVote)
                                      throws DistributedWekaException

Aggregated the supplied list of classifiers. Might discard some classifiers if they have not seen enough training data.

Parameters:

classifiers - the list of classifiers to aggregate

numTrainingInstancesPerClassifier - a list of integers, where each entry is the number of training instances seen by the corresponding classifier

forceVote - true if a Vote ensemble is to be created (even if all classifiers could be directly aggregated to one model of the same type

Returns:

the aggregated classifier

Throws:

DistributedWekaException - if a problem occurs

getDiscarded

public java.util.List<java.lang.Integer> getDiscarded()

Get list of indices of the classifiers that were discarded (if any)

Returns:

a list of indices of discarded classifiers

setMinTrainingFraction

public void setMinTrainingFraction(double m)

Set the minimum training fraction by which a classifier is discarded. This is a fraction of the largest number of training instances processed by the classifiers to allow a classifier to be aggregated. Default = 50%, so any classifier that has trained on less than 50% of the number of instances processed by the classifier that has seen the most data is discarded.

Parameters:

m - a number between 0 and 1.

getMinTrainingFraction

public double getMinTrainingFraction()

Get the minimum training fraction by which a classifier is discarded. This is a fraction of the largest number of training instances processed by the classifiers to allow a classifier to be aggregated. Default = 50%, so any classifier that has trained on less than 50% of the number of instances processed by the classifier that has seen the most data is discarded.

Returns:

the minumum training fraction