jsat.classifiers.knn

Class DANN

java.lang.Object

jsat.classifiers.knn.DANN

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, Classifier, Parameterized

public class DANN
extends java.lang.Object
implements Classifier, Parameterized

DANN is an implementation of Discriminant Adaptive Nearest Neighbor. DANN has a fixed O(n) classification time. At each classification, DANN uses a large set of points to iteratively create and adjust a distance metic that reflects the separability of classes at a localized level. This increases the work considerably over a normal NearestNeighbour classifier. The localized metric is similar to the MahalanobisDistance
Because DANN builds its own metric, it is not possible to provide one. The VectorCollectionFactory allowed in the constructor is to accelerate the first convergence step. In homogeneous areas of the data set, queries can be answered in O(log n) if the vector collection supports it.

See: Hastie, T.,&Tibshirani, R. (1996). Discriminant adaptive nearest neighbor classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18(6), 607–616. doi:10.1109/34.506411

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static double	DEFAULT_EPS The default regularization used when building a metric is 1.0
static int	DEFAULT_ITERATIONS The default number of iterations for creating the metric is 1
static int	DEFAULT_K The default number of neighbors to use when classifying is 1
static int	DEFAULT_KN The default number of neighbors to use when building a metric is 40.

Constructor Summary

Constructors

Constructor and Description
DANN() Creates a new DANN classifier
DANN(int kn, int k) Creates a new DANN classifier
DANN(int kn, int k, double eps) Creates a new DANN classifier
DANN(int kn, int k, double eps, int maxIterations, VectorCollection<VecPaired<Vec,java.lang.Integer>> vcf) Creates a new DANN classifier
DANN(int kn, int k, double eps, VectorCollection<VecPaired<Vec,java.lang.Integer>> vcf) Creates a new DANN classifier

Method Summary

Modifier and Type	Method and Description
CategoricalResults	classify(DataPoint data) Performs classification on the given data point.
Classifier	clone()
double	getEpsilon() Returns the regularization parameter that is applied to the diagonal of the matrix when creating each new metric.
int	getK() Returns the number of nearest neighbors to use when predicting
int	getKn() Returns the number of nearest neighbors to use when adapting the distance metric
int	getMaxIterations() Returns the number of times the distance metric will be updated.
static Distribution	guessK(DataSet d) Guesses the distribution to use for the number of neighbors to consider
static Distribution	guessKn(DataSet d) Guesses the distribution to use for the number of neighbors to consider
void	setEpsilon(double eps) Sets the regularization to apply the the diagonal of the scatter matrix when creating each new metric.
void	setK(int k) Sets the number of nearest neighbors to use when predicting
void	setKn(int kn) Sets the number of nearest neighbors to use when adapting the distance metric.
void	setMaxIterations(int maxIterations) Sets the number of times a new distance metric will be created for each query.
boolean	supportsWeightedData() Indicates whether the model knows how to train using weighted data points.
void	train(ClassificationDataSet dataSet, boolean parallel) Trains the classifier and constructs a model for classification using the given data set.

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.classifiers.Classifier

train

Methods inherited from interface jsat.parameters.Parameterized

getParameter, getParameters

Field Detail

DEFAULT_KN

public static final int DEFAULT_KN

The default number of neighbors to use when building a metric is 40.

See Also:

Constant Field Values

DEFAULT_K

public static final int DEFAULT_K

The default number of neighbors to use when classifying is 1

See Also:

Constant Field Values

DEFAULT_EPS

public static final double DEFAULT_EPS

The default regularization used when building a metric is 1.0

See Also:

Constant Field Values

DEFAULT_ITERATIONS

public static final int DEFAULT_ITERATIONS

The default number of iterations for creating the metric is 1

See Also:

Constant Field Values

Constructor Detail

DANN

public DANN()

Creates a new DANN classifier

DANN

public DANN(int kn,
            int k)

Creates a new DANN classifier

Parameters:

kn - the number of neighbors to use in casting a net to build a new metric

k - the number of neighbors to use with the final metric in classification

DANN

public DANN(int kn,
            int k,
            double eps)

Creates a new DANN classifier

Parameters:

kn - the number of neighbors to use in casting a net to build a new metric

k - the number of neighbors to use with the final metric in classification

eps - the regularization to add to the metric creation

DANN

public DANN(int kn,
            int k,
            double eps,
            VectorCollection<VecPaired<Vec,java.lang.Integer>> vcf)

Creates a new DANN classifier

Parameters:

kn - the number of neighbors to use in casting a net to build a new metric

k - the number of neighbors to use with the final metric in classification

eps - the regularization to add to the metric creation

vcf - the default vector collection that will be used for initial neighbor search

DANN

public DANN(int kn,
            int k,
            double eps,
            int maxIterations,
            VectorCollection<VecPaired<Vec,java.lang.Integer>> vcf)

Creates a new DANN classifier

Parameters:

kn - the number of neighbors to use in casting a net to build a new metric

k - the number of neighbors to use with the final metric in classification

eps - the regularization to add to the metric creation

maxIterations - the maximum number of times to adjust the metric for each classification

vcf - the default vector collection that will be used for initial neighbor search

Method Detail

setK

public void setK(int k)

Sets the number of nearest neighbors to use when predicting

Parameters:

k - the number of neighbors

getK

public int getK()

Returns the number of nearest neighbors to use when predicting

Returns:

the number of neighbors used for classification

setKn

public void setKn(int kn)

Sets the number of nearest neighbors to use when adapting the distance metric. At each iteration of the algorithm, a new distance metric will be created. A larger number of neighbors is used to create a net of points, around which the metric will be adapted.

Parameters:

kn - the number of neighbors to use

getKn

public int getKn()

Returns the number of nearest neighbors to use when adapting the distance metric

Returns:

the number of neighbors used to adapt the metric

setMaxIterations

public void setMaxIterations(int maxIterations)

Sets the number of times a new distance metric will be created for each query. The metric should converge quickly. For this reason, and do to a lack of performance difference, it is highly recommended to use the default of 1 iteration.

Parameters:

maxIterations - the maximum number of times the metric will be updated

getMaxIterations

public int getMaxIterations()

Returns the number of times the distance metric will be updated.

Returns:

the number of iterations the metric will be updated

setEpsilon

public void setEpsilon(double eps)

Sets the regularization to apply the the diagonal of the scatter matrix when creating each new metric.

Parameters:

eps - the regularization value

getEpsilon

public double getEpsilon()

Returns the regularization parameter that is applied to the diagonal of the matrix when creating each new metric.

Returns:

the regularization used.

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.

train

public void train(ClassificationDataSet dataSet,
                  boolean parallel)

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:

train in interface Classifier

Parameters:

dataSet - the data set to train on

parallel - true if multiple threads should be used to train the model. false if it should be done in a single threaded manner.

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

clone

public Classifier clone()

Specified by:

clone in interface Classifier

Overrides:

clone in class java.lang.Object

guessK

public static Distribution guessK(DataSet d)

Guesses the distribution to use for the number of neighbors to consider

Parameters:

d - the dataset to get the guess for

Returns:

the guess for the K parameter

guessKn

public static Distribution guessKn(DataSet d)

Guesses the distribution to use for the number of neighbors to consider

Parameters:

d - the dataset to get the guess for

Returns:

the guess for the Kn parameter