Package weka.classifiers.lazy.AM.label

Class IntLabeler

java.lang.Object

weka.classifiers.lazy.AM.label.Labeler

weka.classifiers.lazy.AM.label.IntLabeler

public class IntLabeler
extends Labeler

Analogical Modeling uses labels composed of boolean vectors in order to group instances into subcontexts and subcontexts in supracontexts. Training set instances are assigned labels by comparing them with the instance to be classified and encoding matched attributes and mismatched attributes in a boolean vector. For example, if we were classifying an instance <a, b, c>, and we had three training instances <x, y, c>, <w, m, c> and <a, b, z>, and used 'n' to represent mismatches and 'y' for matches, the labels would be <n, n, y>, <n, n, y>, and <y, y, n>. The current implementation takes advantage of binary arithmetic by representing mismatches as a 1 bit and matches as a 0 bit, all packed into a 32-bit integer.

Nested Class Summary

Nested classes/interfaces inherited from class weka.classifiers.lazy.AM.label.Labeler

Labeler.Partition

Constructor Summary

Constructors

Constructor	Description
IntLabeler(weka.core.Instance instance, boolean ignoreUnknowns, MissingDataCompare mdc)

Method Summary

Modifier and Type	Method	Description
Label	fromBits(int labelBits)	For testing purposes, this method allows the client to directly specify the label using the bits of an integer
Label	getLatticeBottom()	Creates and returns the label which belongs at the bottom of the boolean lattice formed by the subcontexts labeled by this labeler, i.e.
Label	getLatticeTop()	Creates and returns the label which belongs at the top of the boolean lattice formed by the subcontexts labeled by this labeler, i.e.
IntLabel	label(weka.core.Instance data)	Create a context label for the input instance by comparing it with the test instance.
Label	partition(Label label, int partitionIndex)	In distributed processing, it is necessary to split labels into partitions.

Methods inherited from class weka.classifiers.lazy.AM.label.Labeler

getCardinality, getCardinality, getContextList, getContextString, getIgnoreUnknowns, getInstanceAttNamesList, getInstanceAttsString, getInstanceAttValuesList, getMissingDataCompare, getTestInstance, isIgnored, numPartitions

Methods inherited from class java.lang.Object

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

Constructor Detail

IntLabeler

public IntLabeler(weka.core.Instance instance,
                  boolean ignoreUnknowns,
                  MissingDataCompare mdc)

Throws:

java.lang.IllegalArgumentException - if the cardinality of the input instance is greater than 32

Method Detail

label

public IntLabel label(weka.core.Instance data)

Description copied from class: Labeler

Create a context label for the input instance by comparing it with the test instance.

Specified by:

label in class Labeler

Parameters:

data - Instance to be labeled

Returns:

the label for the context that the instance belongs to. The cardinality of the label will be the same as the test and data items. At any given index i, label.matches(i) will return true if that feature is the same in the test and data instances.

getLatticeTop

public Label getLatticeTop()

Description copied from class: Labeler

Creates and returns the label which belongs at the top of the boolean lattice formed by the subcontexts labeled by this labeler, i.e. the one for which every feature is a match.

Specified by:

getLatticeTop in class Labeler

Returns:

A label with all matches

getLatticeBottom

public Label getLatticeBottom()

Description copied from class: Labeler

Creates and returns the label which belongs at the bottom of the boolean lattice formed by the subcontexts labeled by this labeler, i.e. the one for which every feature is a mismatch.

Specified by:

getLatticeBottom in class Labeler

Returns:

A label with all mismatches

fromBits

public Label fromBits(int labelBits)

Description copied from class: Labeler

For testing purposes, this method allows the client to directly specify the label using the bits of an integer

Specified by:

fromBits in class Labeler

partition

public Label partition(Label label,
                       int partitionIndex)

Description copied from class: Labeler

In distributed processing, it is necessary to split labels into partitions. This method returns a partition for the given label. A full label is partitioned into pieces 0 through Labeler.numPartitions(), so code to process labels in pieces should look like this:

        Label myLabel = myLabeler.label(myInstance);
        for(int i = 0; i < myLabeler.numPartitions(); i++)
                process(myLabeler.partition(myLabel, i);
 

Specified by:

partition in class Labeler

partitionIndex - index of the partition to return

Returns:

a new label representing a portion of the attributes represented by the input label.