jsat.distributions.kernels

Class KernelPoint

java.lang.Object

jsat.distributions.kernels.KernelPoint

public class KernelPoint
extends java.lang.Object

The Kernel Point represents a kernelized weight vector by a linear combination of vectors transformed through a kernel fuctiion. This implementation allows the selection of multiple different budget maintenance strategies

See KernelRLS and Projectron for methods and papers based on the same ideas used to create this class.
Credit goes to Davis King of the dlib library for the idea of this type of class.

Changing the budget maintinance method or other parameters should be done before adding any data points to the KernelPoint.
If a maximum budget is specified, it may always be increased - but may not be decreased.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	KernelPoint.BudgetStrategy These enums control the method used to reduce the size of the support vector set in the kernel point.

Constructor Summary

Constructors

Constructor and Description
KernelPoint(KernelPoint toCopy) Copy constructor
KernelPoint(KernelTrick k, double errorTolerance) Creates a new Kernel Point, which is a point in the kernel space represented by an accumulation of vectors and uses the KernelPoint.BudgetStrategy.PROJECTION strategy with an unbounded maximum budget

Method Summary

Modifier and Type	Method and Description
KernelPoint	clone()
double	dist(KernelPoint x) Computes the Euclidean distance between this kernel point and the given kernel point in the kernel space
double	dist(Vec x) Computes the Euclidean distance between this kernel point and the given input in the kernel space
double	dist(Vec x, java.util.List<java.lang.Double> qi) Computes the Euclidean distance between this kernel point and the given input in the kernel space
double	dot(KernelPoint x) Returns the dot product between this point and another in the kernel space
double	dot(Vec x) Computes the dot product between the kernel point this object represents and the given input vector in the kernel space.
double	dot(Vec x, java.util.List<java.lang.Double> qi) Computes the dot product between the kernel point this object represents and the given input vector in the kernel space
int	getBasisSize() Returns the number of vectors serving as the basis set
KernelPoint.BudgetStrategy	getBudgetStrategy() Returns the budget method used
double	getErrorTolerance() Returns the error tolerance that is used depending on the KernelPoint.BudgetStrategy in use
int	getMaxBudget() Returns the current maximum budget for support vectors
java.util.List<Vec>	getRawBasisVecs() Returns the list of the raw vectors being used by the kernel points.
double	getSqrdNorm() Returns the squared values of the 2 norm of the point this object represents
void	mutableAdd(double c, Vec x_t) Alters this point to contain the given input vector as well
void	mutableAdd(double c, Vec x_t, java.util.List<java.lang.Double> qi) Alters this point to contain the given input vector as well
void	mutableAdd(Vec x_t) Alters this point to contain the given input vector as well
void	mutableMultiply(double c) Alters this point to be multiplied by the given value
void	setBudgetStrategy(KernelPoint.BudgetStrategy budgetStrategy) Sets the method used for maintaining the budget of support vectors.
void	setErrorTolerance(double errorTolerance) Sets the error tolerance used for projection maintenance strategies such as KernelPoint.BudgetStrategy.PROJECTION
void	setMaxBudget(int maxBudget) Sets the maximum budget for support vectors to allow.

Methods inherited from class java.lang.Object

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

Constructor Detail

KernelPoint

public KernelPoint(KernelTrick k,
                   double errorTolerance)

Creates a new Kernel Point, which is a point in the kernel space represented by an accumulation of vectors and uses the KernelPoint.BudgetStrategy.PROJECTION strategy with an unbounded maximum budget

Parameters:

k - the kernel to use

errorTolerance - the maximum error in [0, 1] allowed for projecting a vector instead of adding it to the basis set

KernelPoint

public KernelPoint(KernelPoint toCopy)

Copy constructor

Parameters:

toCopy - the object to copy

Method Detail

setMaxBudget

public void setMaxBudget(int maxBudget)

Sets the maximum budget for support vectors to allow. Setting to Integer.MAX_VALUE is essentially an unbounded number of support vectors. Increasing the budget after adding the first vector is always allowed, but it may not be possible to reduce the number of current support vectors is above the desired budget.

Parameters:

maxBudget - the maximum number of allowed support vectors

getMaxBudget

public int getMaxBudget()

Returns the current maximum budget for support vectors

Returns:

the maximum budget for support vectors

setBudgetStrategy

public void setBudgetStrategy(KernelPoint.BudgetStrategy budgetStrategy)

Sets the method used for maintaining the budget of support vectors. This method must be called before any vectors are added to the KernelPoint.

The budget maintenance strategy used controls the time complexity and memory use of the model.

Parameters:

budgetStrategy - the budget maintenance strategy

getBudgetStrategy

public KernelPoint.BudgetStrategy getBudgetStrategy()

Returns the budget method used

Returns:

the budget method used

setErrorTolerance

public void setErrorTolerance(double errorTolerance)

Sets the error tolerance used for projection maintenance strategies such as KernelPoint.BudgetStrategy.PROJECTION

Parameters:

errorTolerance - the error tolerance in [0, 1]

getErrorTolerance

public double getErrorTolerance()

Returns the error tolerance that is used depending on the KernelPoint.BudgetStrategy in use

Returns:

the error tolerance value

getSqrdNorm

public double getSqrdNorm()

Returns the squared values of the 2 norm of the point this object represents

Returns:

the squared value of the 2 norm

dot

public double dot(Vec x)

Computes the dot product between the kernel point this object represents and the given input vector in the kernel space.

Parameters:

x - the input vector to work with

Returns:

the dot product in the kernel space between this point and x

dot

public double dot(Vec x,
                  java.util.List<java.lang.Double> qi)

Computes the dot product between the kernel point this object represents and the given input vector in the kernel space

Parameters:

x - the input vector to work with

qi - the query information for the vector, or null only if the kernel in use does not support acceleration.

Returns:

the dot product in the kernel space between this point and x

dot

public double dot(KernelPoint x)

Returns the dot product between this point and another in the kernel space

Parameters:

x - the point to take the dot product with

Returns:

the dot product in the kernel space between this point and x

dist

public double dist(Vec x)

Computes the Euclidean distance between this kernel point and the given input in the kernel space

Parameters:

x - the input vector to work with

Returns:

the Euclidean distance between this point and x in the kernel space

dist

public double dist(Vec x,
                   java.util.List<java.lang.Double> qi)

Computes the Euclidean distance between this kernel point and the given input in the kernel space

Parameters:

x - the input vector to work with

qi - the query information for the vector, or null only if the kernel in use does not support acceleration.

Returns:

the Euclidean distance between this point and x in the kernel space

dist

public double dist(KernelPoint x)

Computes the Euclidean distance between this kernel point and the given kernel point in the kernel space

Parameters:

x - the input point to work with

Returns:

the Euclidean distance between this point and x in the kernel space

mutableMultiply

public void mutableMultiply(double c)

Alters this point to be multiplied by the given value

Parameters:

c - the value to multiply by

mutableAdd

public void mutableAdd(Vec x_t)

Alters this point to contain the given input vector as well

Parameters:

x_t - the vector to add

mutableAdd

public void mutableAdd(double c,
                       Vec x_t)

Alters this point to contain the given input vector as well

Parameters:

c - the multiplicative constant to apply with the vector

x_t - the vector to add

mutableAdd

public void mutableAdd(double c,
                       Vec x_t,
                       java.util.List<java.lang.Double> qi)

Alters this point to contain the given input vector as well

Parameters:

c - the multiplicative constant to apply with the vector

x_t - the vector to add

qi - the query information for the vector, or null only if the kernel in use does not support acceleration.

getBasisSize

public int getBasisSize()

Returns the number of vectors serving as the basis set

Returns:

the number of vectors serving as the basis set

getRawBasisVecs

public java.util.List<Vec> getRawBasisVecs()

Returns the list of the raw vectors being used by the kernel points. Altering this vectors will alter the same vectors used by the KernelPoint and will cause inconsistent results.

The returned list may not be modified

Returns:

a the list of all the vectors in use as a basis set by this KernelPoint

clone

public KernelPoint clone()

Overrides:

clone in class java.lang.Object