jsat.datatransform

Class WhitenedPCA

java.lang.Object

jsat.datatransform.DataTransformBase

jsat.datatransform.WhitenedPCA

All Implemented Interfaces:

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

Direct Known Subclasses:

WhitenedZCA

public class WhitenedPCA
extends DataTransformBase

An extension of PCA that attempts to capture the variance, and make the variables in the output space independent from each-other. An of equal scale, so that the covariance is equal to I. The results may be further from the identity matrix than desired as the target dimension shrinks

The Whitened PCA is more computational expensive than the normal PCA algorithm, but transforming the data takes the same time.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
WhitenedPCA() Creates a new WhitenedPCA transform that uses up to 50 dimensions for the transformed space.
WhitenedPCA(DataSet dataSet) Creates a new WhitenedPCA.
WhitenedPCA(DataSet dataSet, double regularization) Creates a new WhitenedPCA, the dimensions will be chosen so that the subset of dimensions is of full rank.
WhitenedPCA(DataSet dataSet, double regularization, int dims) Creates a new WhitenedPCA from the given dataset
WhitenedPCA(DataSet dataSet, int dims) Creates a new WhitenedPCA.
WhitenedPCA(double regularization, int dims) Creates a new WhitenedPCA transform
WhitenedPCA(int dims) Creates a new WhitenedPCA transform

Method Summary

Modifier and Type	Method and Description
DataTransform	clone()
void	fit(DataSet dataSet) Fits this transform to the given dataset.
int	getDimensions() Returns the number of dimensions to project down to
double	getRegularization()
static Distribution	guessDimensions(DataSet d)
void	setDimensions(int dimensions) Sets the number of dimensions to project down to
void	setRegularization(double regularization)
DataPoint	transform(DataPoint dp) Returns a new data point that is a transformation of the original data point.

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.parameters.Parameterized

getParameter, getParameters

Constructor Detail

WhitenedPCA

public WhitenedPCA()

Creates a new WhitenedPCA transform that uses up to 50 dimensions for the transformed space. This may not be optimal for any given dataset.

Parameters:

dims - the number of dimensions to project down to

WhitenedPCA

public WhitenedPCA(int dims)

Creates a new WhitenedPCA transform

Parameters:

dims - the number of dimensions to project down to

WhitenedPCA

public WhitenedPCA(double regularization,
                   int dims)

Creates a new WhitenedPCA transform

Parameters:

regularization - the amount of regularization to add, avoids numerical instability

dims - the number of dimensions to project down to

WhitenedPCA

public WhitenedPCA(DataSet dataSet,
                   double regularization,
                   int dims)

Creates a new WhitenedPCA from the given dataset

Parameters:

dataSet - the data set to whiten

regularization - the amount of regularization to add, avoids numerical instability

dims - the number of dimensions to project down to

WhitenedPCA

public WhitenedPCA(DataSet dataSet,
                   double regularization)

Creates a new WhitenedPCA, the dimensions will be chosen so that the subset of dimensions is of full rank.

Parameters:

dataSet - the data set to whiten

regularization - the amount of regularization to add, avoids numerical instability

WhitenedPCA

public WhitenedPCA(DataSet dataSet)

Creates a new WhitenedPCA. The dimensions will be chosen so that the subset of dimensions is of full rank. The regularization parameter will be chosen as the log of the condition of the covariance.

Parameters:

dataSet - the data set to whiten

WhitenedPCA

public WhitenedPCA(DataSet dataSet,
                   int dims)

Creates a new WhitenedPCA. The regularization parameter will be chosen as the log of the condition of the covariance.

Parameters:

dataSet - the data set to whiten

dims - the number of dimensions to project down to

Method Detail

fit

public void fit(DataSet dataSet)

Description copied from interface: DataTransform

Fits this transform to the given dataset. Some transforms can only be learned from classification or regression datasets. If an incompatible dataset type is given, a FailedToFitException exception may be thrown.

Parameters:

dataSet - the dataset to fir this transform to

transform

public DataPoint transform(DataPoint dp)

Description copied from interface: DataTransform

Returns a new data point that is a transformation of the original data point. This new data point is a different object, but may contain the same references as the original data point. It is not guaranteed that you can mutate the transformed point without having a side effect on the original point.

Parameters:

dp - the data point to apply a transformation to

Returns:

a transformed data point

setRegularization

public void setRegularization(double regularization)

Parameters:

regularization - the regularization to apply to the diagonal of the decomposition. This can improve numeric stability and reduces noise.

getRegularization

public double getRegularization()

Returns:

the amount of regularization to apply

clone

public DataTransform clone()

Specified by:

clone in interface DataTransform

Specified by:

clone in class DataTransformBase

setDimensions

public void setDimensions(int dimensions)

Sets the number of dimensions to project down to

Parameters:

dimensions - the feature size to project down to

getDimensions

public int getDimensions()

Returns the number of dimensions to project down to

Returns:

the number of dimensions to project down to

guessDimensions

public static Distribution guessDimensions(DataSet d)