jsat.datatransform.kernel

Class RFF_RBF

java.lang.Object

jsat.datatransform.DataTransformBase

jsat.datatransform.kernel.RFF_RBF

All Implemented Interfaces:

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

public class RFF_RBF
extends DataTransformBase

An Implementation of Random Fourier Features for the RBFKernel. It transforms the numerical variables of a feature space to form a new feature space where the dot product between features approximates the RBF Kernel product.

See: Rahimi, A.,&Recht, B. (2007). Random Features for Large-Scale Kernel Machines. Neural Information Processing Systems. Retrieved from here

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
RFF_RBF() Creates a new RFF RBF object that will use an transformed feature space with a dimensionality of 512.
RFF_RBF(double sigma) Creates a new RFF RBF object that will use an transformed feature space with a dimensionality of 512.
RFF_RBF(double sigma, int dim) Creates a new RFF RBF object
RFF_RBF(double sigma, int dim, boolean inMemory) Creates a new RFF RBF object
RFF_RBF(int featurSize, double sigma, int dim, java.util.Random rand, boolean inMemory) Creates a new RFF RBF object

Method Summary

Modifier and Type	Method and Description
RFF_RBF	clone()
void	fit(DataSet data) Fits this transform to the given dataset.
int	getDimensions() Returns the number of dimensions that will be used in the projected space
double	getSigma() Returns the σ value used for the RBF kernel approximation.
Distribution	guessSigma(DataSet d) Guess the distribution to use for the kernel width term σ in the RBF kernel being approximated.
boolean	isInMemory()
void	setDimensions(int dimensions) Sets the number of dimensions in the new approximate space to use.
void	setInMemory(boolean inMemory) Sets whether or not the transform matrix is stored explicitly in memory or not.
void	setSigma(double sigma) Sets the σ parameter of the RBF kernel that is being approximated.
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

RFF_RBF

public RFF_RBF()

Creates a new RFF RBF object that will use an transformed feature space with a dimensionality of 512. This constructor should be used with a parameter search to find a good value for sigma

RFF_RBF

public RFF_RBF(double sigma)

Creates a new RFF RBF object that will use an transformed feature space with a dimensionality of 512.

Parameters:

sigma - the positive sigma value for the RBFKernel

RFF_RBF

public RFF_RBF(double sigma,
               int dim)

Creates a new RFF RBF object

Parameters:

sigma - the positive sigma value for the RBFKernel

dim - the new feature size dimension to project into.

RFF_RBF

public RFF_RBF(double sigma,
               int dim,
               boolean inMemory)

Creates a new RFF RBF object

Parameters:

sigma - the positive sigma value for the RBFKernel

dim - the new feature size dimension to project into.

inMemory - true if the internal matrix should be stored in memory. If false, the memory will be re-computed as needed, increasing computation cost but uses no extra memory.

RFF_RBF

public RFF_RBF(int featurSize,
               double sigma,
               int dim,
               java.util.Random rand,
               boolean inMemory)

Creates a new RFF RBF object

Parameters:

featurSize - the number of numeric features in the original feature space

sigma - the positive sigma value for the RBFKernel

dim - the new feature size dimension to project into.

rand - the source of randomness to initialize internal state

inMemory - true if the internal matrix should be stored in memory. If false, the memory will be re-computed as needed, increasing computation cost but uses no extra memory.

Method Detail

fit

public void fit(DataSet data)

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:

data - 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

clone

public RFF_RBF clone()

Specified by:

clone in interface DataTransform

Specified by:

clone in class DataTransformBase

setInMemory

public void setInMemory(boolean inMemory)

Sets whether or not the transform matrix is stored explicitly in memory or not. Explicit storage is often faster, but can be prohibitive for large feature sizes

Parameters:

inMemory - true to explicitly store the transform matrix, false to re-create it on the fly as needed

isInMemory

public boolean isInMemory()

Returns:

true if this object will explicitly store the transform matrix, false to re-create it on the fly as needed

setDimensions

public void setDimensions(int dimensions)

Sets the number of dimensions in the new approximate space to use. This will be the number of numeric features in the transformed data, and larger values increase the accuracy of the approximation.

Parameters:

dimensions -

getDimensions

public int getDimensions()

Returns the number of dimensions that will be used in the projected space

Returns:

the number of dimensions that will be used in the projected space

setSigma

public void setSigma(double sigma)

Sets the σ parameter of the RBF kernel that is being approximated.

Parameters:

sigma - the positive value to use for σ

See Also:

RBFKernel.setSigma(double)

getSigma

public double getSigma()

Returns the σ value used for the RBF kernel approximation.

Returns:

the σ value used for the RBF kernel approximation.

guessSigma

public Distribution guessSigma(DataSet d)

Guess the distribution to use for the kernel width term σ in the RBF kernel being approximated.

Parameters:

d - the data set to get the guess for

Returns:

the guess for the σ parameter in the RBF Kernel