jsat.classifiers.linear

Class AROW

java.lang.Object

jsat.classifiers.BaseUpdateableClassifier

jsat.classifiers.linear.AROW

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, BinaryScoreClassifier, Classifier, UpdateableClassifier, Parameterized, SimpleWeightVectorModel, SingleWeightVectorModel

public class AROW
extends BaseUpdateableClassifier
implements BinaryScoreClassifier, Parameterized, SingleWeightVectorModel

An implementation of Adaptive Regularization of Weight Vectors (AROW), which uses second order information to learn a large margin binary classifier. As such, updates can occur on correctly classified instances if they are not far enough from the margin. Unlike many margin algorithms, it handles noise well.
NOTE: This implementation does not add an implicit bias term, so the solution goes through the origin

See: Crammer, K., Kulesza, A.,&Dredze, M. (2013). Adaptive regularization of weight vectors. Machine Learning, 91(2), 155–187. doi:10.1007/s10994-013-5327-x

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
AROW() Creates a new AROW learner
AROW(double r, boolean diagonalOnly) Creates a new AROW learner

Method Summary

Modifier and Type	Method and Description
CategoricalResults	classify(DataPoint data) Performs classification on the given data point.
AROW	clone()
double	getBias() Returns the bias term used for the model, or 0 of the model does not support or was not trained with a bias term.
double	getBias(int index) Returns the bias term used with the weight vector for the given class index.
double	getR() Returns the regularization parameter
Vec	getRawWeight() Returns the only weight vector used for the model
Vec	getRawWeight(int index) Returns the raw weight vector associated with the given class index.
double	getScore(DataPoint dp) Returns the numeric score for predicting a class of a given data point, where the sign of the value indicates which class the data point is predicted to belong to.
Vec	getWeightVec() Returns the weight vector used to compute results via a dot product.
static Distribution	guessR(DataSet d) Guess the distribution to use for the regularization term r .
boolean	isDiagonalOnly() Returns true if the covariance matrix is restricted to its diagonal entries
int	numWeightsVecs() Returns the number of weight vectors that can be returned.
void	setDiagonalOnly(boolean diagonalOnly) Using the full covariance matrix requires O(d2) work on mistakes, where d is the dimension of the data.
void	setR(double r) Sets the r parameter of AROW, which controls the regularization.
void	setUp(CategoricalData[] categoricalAttributes, int numericAttributes, CategoricalData predicting) Prepares the classifier to begin learning from its UpdateableClassifier.update(jsat.classifiers.DataPoint, int) method.
boolean	supportsWeightedData() Indicates whether the model knows how to train using weighted data points.
void	update(DataPoint dataPoint, int targetClass) Updates the classifier by giving it a new data point to learn from.

Methods inherited from class jsat.classifiers.BaseUpdateableClassifier

getEpochs, setEpochs, train, train, trainEpochs

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.classifiers.Classifier

train, train

Methods inherited from interface jsat.parameters.Parameterized

getParameter, getParameters

Constructor Detail

AROW

public AROW()

Creates a new AROW learner

AROW

public AROW(double r,
            boolean diagonalOnly)

Creates a new AROW learner

Parameters:

r - the regularization parameter

diagonalOnly - whether or not to use only the diagonal of the covariance

See Also:

setR(double), setDiagonalOnly(boolean)

Method Detail

setDiagonalOnly

public void setDiagonalOnly(boolean diagonalOnly)

Using the full covariance matrix requires O(d²) work on mistakes, where d is the dimension of the data. Runtime can be reduced by using only the diagonal of the matrix to perform updates in O(s) time, where s ≤ d is the number of non-zero values in the input

Parameters:

diagonalOnly - true to use only the diagonal of the covariance

isDiagonalOnly

public boolean isDiagonalOnly()

Returns true if the covariance matrix is restricted to its diagonal entries

Returns:

true if the covariance matrix is restricted to its diagonal entries

setR

public void setR(double r)

Sets the r parameter of AROW, which controls the regularization. Larger values reduce the change in the model on each update.

Parameters:

r - the regularization parameter in (0, Inf)

getR

public double getR()

Returns the regularization parameter

Returns:

the regularization parameter

getWeightVec

public Vec getWeightVec()

Returns the weight vector used to compute results via a dot product.
Do not modify this value, or you will alter the results returned.

Returns:

the learned weight vector for prediction

clone

public AROW clone()

Specified by:

clone in interface BinaryScoreClassifier

Specified by:

clone in interface Classifier

Specified by:

clone in interface UpdateableClassifier

Specified by:

clone in class BaseUpdateableClassifier

setUp

public void setUp(CategoricalData[] categoricalAttributes,
                  int numericAttributes,
                  CategoricalData predicting)

Description copied from interface: UpdateableClassifier

Prepares the classifier to begin learning from its UpdateableClassifier.update(jsat.classifiers.DataPoint, int) method.

Specified by:

setUp in interface UpdateableClassifier

Parameters:

categoricalAttributes - an array containing the categorical attributes that will be in each data point

numericAttributes - the number of numeric attributes that will be in each data point

predicting - the information for the target class that will be predicted

update

public void update(DataPoint dataPoint,
                   int targetClass)

Description copied from interface: UpdateableClassifier

Updates the classifier by giving it a new data point to learn from.

Specified by:

update in interface UpdateableClassifier

Parameters:

dataPoint - the data point to learn

targetClass - the target class of the data point

classify

public CategoricalResults classify(DataPoint data)

Description copied from interface: Classifier

Performs classification on the given data point.

Specified by:

classify in interface Classifier

Parameters:

data - the data point to classify

Returns:

the results of the classification.

getScore

public double getScore(DataPoint dp)

Description copied from interface: BinaryScoreClassifier

Returns the numeric score for predicting a class of a given data point, where the sign of the value indicates which class the data point is predicted to belong to.

Specified by:

getScore in interface BinaryScoreClassifier

Parameters:

dp - the data point to predict the class label of

Returns:

the score for the given data point

supportsWeightedData

public boolean supportsWeightedData()

Description copied from interface: Classifier

Indicates whether the model knows how to train using weighted data points. If it does, the model will train assuming the weights. The values returned by this method may change depending on the parameters set for the model.

Specified by:

supportsWeightedData in interface Classifier

Returns:

true if the model supports weighted data, false otherwise

getRawWeight

public Vec getRawWeight()

Description copied from interface: SingleWeightVectorModel

Returns the only weight vector used for the model

Specified by:

getRawWeight in interface SingleWeightVectorModel

Returns:

the only weight vector used for the model

getBias

public double getBias()

Description copied from interface: SingleWeightVectorModel

Returns the bias term used for the model, or 0 of the model does not support or was not trained with a bias term.

Specified by:

getBias in interface SingleWeightVectorModel

Returns:

the bias term for the model

getRawWeight

public Vec getRawWeight(int index)

Description copied from interface: SimpleWeightVectorModel

Returns the raw weight vector associated with the given class index. If the given class is an implicit zero vector, a ConstantVector object may be returned.
Do not alter the returned weight vector, as it will change the model's values.

If a regression problem, only index = 0 should be used

Specified by:

getRawWeight in interface SimpleWeightVectorModel

Parameters:

index - the class index to get the weight vector for

Returns:

the weight vector used for the specified class

getBias

public double getBias(int index)

Description copied from interface: SimpleWeightVectorModel

Returns the bias term used with the weight vector for the given class index. If the model does not support or was not trained with bias weights, 0 will be returned.

If a regression problem, only index = 0 should be used

Specified by:

getBias in interface SimpleWeightVectorModel

Parameters:

index - the class index to get the weight vector for

Returns:

the bias term for the specified class

numWeightsVecs

public int numWeightsVecs()

Description copied from interface: SimpleWeightVectorModel

Returns the number of weight vectors that can be returned. For binary classification problems the value may be 1 if only a single weight vector's sign is used to determine the class. For multi-class problems, the weight vector count includes the implicit zero vector (if one is being used).

Specified by:

numWeightsVecs in interface SimpleWeightVectorModel

Returns:

the number of weight vectors for which SimpleWeightVectorModel.getRawWeight(int) can be called.

guessR

public static Distribution guessR(DataSet d)

Guess the distribution to use for the regularization term r .

Parameters:

d - the data set to get the guess for

Returns:

the guess for the r parameter