jsat.classifiers.neuralnetwork

Class LVQLLC

java.lang.Object

jsat.classifiers.neuralnetwork.LVQ

jsat.classifiers.neuralnetwork.LVQLLC

All Implemented Interfaces:

Serializable, Cloneable, Classifier, Parameterized

public class LVQLLC
extends LVQ

LVQ with Locally Learned Classifier (LVQ-LLC) is an adaption of the LVQ algorithm I have come up with. Given a classification data set, LVQ develops and moves several prototype vectors throughout the space, trying to place them as good representatives. Classification is then done Nearest Neighbor style among the prototypes.
LVQ-LLC trains a local classifier on all of the data points that belong to the prototype, and the data points that lie across the border but are still near the prototype using the LVQ.getEpsilonDistance() parameter that is used to update two prototypes at the same time. Classification can then be done by getting the Classifier for the nearest prototype, or averaging the results of the two closest prototypes if the point is near a boundary.
This is done because, given a complex decision boundary and a sufficient number of prototypes, each prototype's domain will be a smaller subset of the problem and will hopefully resemble a simpler decision problem that can be solved by a less complicated local learner.
LVQ-LLC has the following advantages over LVQ:

• Can return probabilities instead of hard classifications
• Approximate decision boundaries can be more complicated than voronoi diagrams
• Increase accuracy given a smaller number of prototypes per class

By default, the local classifier is the MultivariateNormals.

Author:

Edward Raff

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class jsat.classifiers.neuralnetwork.LVQ

LVQ.LVQVersion

Field Summary

Fields inherited from class jsat.classifiers.neuralnetwork.LVQ

DEFAULT_EPS, DEFAULT_ITERATIONS, DEFAULT_LEARNING_RATE, DEFAULT_LVQ_METHOD, DEFAULT_MSCALE, DEFAULT_REPS_PER_CLASS, DEFAULT_SEED_SELECTION, DEFAULT_STOPPING_DIST, dm, vc, weightClass, weights, wins

Constructor Summary

Constructors

Modifier	Constructor and Description
	LVQLLC(DistanceMetric dm, int iterations) Creates a new LVQ-LLC instance that uses MultivariateNormals as the local classifier.
	LVQLLC(DistanceMetric dm, int iterations, Classifier localClasifier) Creates a new LVQ-LLC instance
	LVQLLC(DistanceMetric dm, int iterations, Classifier localClasifier, double learningRate, int representativesPerClass) Creates a new LVQ-LLC instance
	LVQLLC(DistanceMetric dm, int iterations, Classifier localClasifier, double learningRate, int representativesPerClass, LVQ.LVQVersion lvqVersion, DecayRate learningDecay) Creates a new LVQ-LLC instance
protected	LVQLLC(LVQLLC toCopy)

Method Summary

Modifier and Type	Method and Description
CategoricalResults	classify(DataPoint data) Performs classification on the given data point.
LVQLLC	clone()
Classifier	getLocalClassifier() Returns the classifier used for each prototype
void	setLocalClassifier(Classifier localClassifier) Each prototype will create a classifier that is local to itself, and trained on the points that belong to the prototype and those near the border of the prototype.
void	trainC(ClassificationDataSet dataSet) Trains the classifier and constructs a model for classification using the given data set.
void	trainC(ClassificationDataSet dataSet, ExecutorService threadPool) Trains the classifier and constructs a model for classification using the given data set.

Methods inherited from class jsat.classifiers.neuralnetwork.LVQ

epsClose, getDistanceMetric, getEpsilonDistance, getIterations, getLearningDecay, getLearningRate, getLVQMethod, getMScale, getParameter, getParameters, getRepresentativesPerClass, getSeedSelection, getStoppingDist, setDistanceMetric, setEpsilonDistance, setIterations, setLearningDecay, setLearningRate, setLVQMethod, setMScale, setRepresentativesPerClass, setSeedSelection, setStoppingDist, setVecCollectionFactory, supportsWeightedData

Methods inherited from class java.lang.Object

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

Constructor Detail

LVQLLC

public LVQLLC(DistanceMetric dm,
              int iterations)

Creates a new LVQ-LLC instance that uses MultivariateNormals as the local classifier.

Parameters:

dm - the distance metric to use

iterations - the number of iterations to perform

LVQLLC

public LVQLLC(DistanceMetric dm,
              int iterations,
              Classifier localClasifier)

Creates a new LVQ-LLC instance

Parameters:

dm - the distance metric to use

iterations - the number of iterations to perform

localClasifier - the classifier to use as a local classifier for each prototype

LVQLLC

public LVQLLC(DistanceMetric dm,
              int iterations,
              Classifier localClasifier,
              double learningRate,
              int representativesPerClass)

Creates a new LVQ-LLC instance

Parameters:

dm - the distance metric to use

iterations - the number of iterations to perform

localClasifier - the classifier to use as a local classifier for each prototype

learningRate - the learning rate to use when updating

representativesPerClass - the number of representatives to create for each class

LVQLLC

public LVQLLC(DistanceMetric dm,
              int iterations,
              Classifier localClasifier,
              double learningRate,
              int representativesPerClass,
              LVQ.LVQVersion lvqVersion,
              DecayRate learningDecay)

Creates a new LVQ-LLC instance

Parameters:

dm - the distance metric to use

iterations - the number of iterations to perform

localClasifier - the classifier to use as a local classifier for each prototype

learningRate - the learning rate to use when updating

representativesPerClass - the number of representatives to create for each class

lvqVersion - the version of LVQ to use

learningDecay - the amount of decay to apply to the learning rate

LVQLLC

protected LVQLLC(LVQLLC toCopy)

Method Detail

setLocalClassifier

public void setLocalClassifier(Classifier localClassifier)

Each prototype will create a classifier that is local to itself, and trained on the points that belong to the prototype and those near the border of the prototype. This sets the classifier that will be used

Parameters:

localClassifier - the local classifier to use for each prototype

getLocalClassifier

public Classifier getLocalClassifier()

Returns the classifier used for each prototype

Returns:

the classifier used for each prototype

classify

public CategoricalResults classify(DataPoint data)

Description copied from interface: Classifier

Performs classification on the given data point.

Specified by:

classify in interface Classifier

Overrides:

classify in class LVQ

Parameters:

data - the data point to classify

Returns:

the results of the classification.

trainC

public void trainC(ClassificationDataSet dataSet,
                   ExecutorService threadPool)

Description copied from interface: Classifier

Trains the classifier and constructs a model for classification using the given data set. If the training method knows how, it will used the threadPool to conduct training in parallel. This method will block until the training has completed.

Specified by:

trainC in interface Classifier

Overrides:

trainC in class LVQ

Parameters:

dataSet - the data set to train on

threadPool - the source of threads to use.

trainC

public void trainC(ClassificationDataSet dataSet)

Description copied from interface: Classifier

Trains the classifier and constructs a model for classification using the given data set.

Specified by:

trainC in interface Classifier

Overrides:

trainC in class LVQ

Parameters:

dataSet - the data set to train on

clone

public LVQLLC clone()

Specified by:

clone in interface Classifier

Overrides:

clone in class LVQ