jsat.classifiers.linear.kernelized

Enum CSKLR.UpdateMode

java.lang.Object

java.lang.Enum<CSKLR.UpdateMode>

jsat.classifiers.linear.kernelized.CSKLR.UpdateMode

All Implemented Interfaces:

Serializable, Comparable<CSKLR.UpdateMode>

Enclosing class:

CSKLR

public static enum CSKLR.UpdateMode
extends Enum<CSKLR.UpdateMode>

Controls when updates are performed on the model. Depending on which update model is used, the acceptable values and behaviors of CSKLR.setEta(double) and CSKLR.setGamma(double) may change.

The "auxiliary" modes perform updates with a probability of log(1+e^-z) / a(z), where a(z) is the auxiliary function and z is the raw margin value times the class label.

Enum Constant Summary

Enum Constants

Enum Constant and Description
AUXILIARY_1 Performs model updates based on a "auxiliary" function a(z) = log(γ + e-z).
AUXILIARY_2 Performs model updates based on a "auxiliary" function a(z) = log(1 + γ e-z).
AUXILIARY_3 Performs model updates based on a "auxiliary" function a(z) = max(loss(z), loss(γ).
MARGIN Performs model updates probabilistically based on their distance from the margin of the classifier.
NC NC stands for Non-Conservative, this mode will perform a model update on every new input, creating a very dense model.

Method Summary

Modifier and Type	Method and Description
protected abstract double	grad(double y, double score, double preScore, double gamma) Get the gradient value that should be applied based on the input variable from the current model
protected abstract double	pt(double y, double score, double preScore, double eta, double gamma) Returns the Bernoulli trial probability variable
static CSKLR.UpdateMode	valueOf(String name) Returns the enum constant of this type with the specified name.
static CSKLR.UpdateMode[]	values() Returns an array containing the constants of this enum type, in the order they are declared.

Methods inherited from class java.lang.Enum

clone, compareTo, equals, finalize, getDeclaringClass, hashCode, name, ordinal, toString, valueOf

Methods inherited from class java.lang.Object

getClass, notify, notifyAll, wait, wait, wait

Enum Constant Detail

NC

public static final CSKLR.UpdateMode NC

NC stands for Non-Conservative, this mode will perform a model update on every new input, creating a very dense model. The learning rate may take on any positive value, and CSKLR.setGamma(double) is not used.

MARGIN

public static final CSKLR.UpdateMode MARGIN

Performs model updates probabilistically based on their distance from the margin of the classifier. In this case, CSKLR.setEta(double) should be less than 2, or the model will become dense. CSKLR.setGamma(double) is not used.

AUXILIARY_1

public static final CSKLR.UpdateMode AUXILIARY_1

Performs model updates based on a "auxiliary" function a(z) = log(γ + e^-z). gamma should be in the range (1, Infinity) where larger values increase the sparsity of the model

This is the main auxiliary method used by the authors. They use values for γ in the range of 1+10^{± x} ∀ x ∈ {0, 1, 2, 3, 4}

AUXILIARY_2

public static final CSKLR.UpdateMode AUXILIARY_2

Performs model updates based on a "auxiliary" function a(z) = log(1 + γ e^-z). gamma should be in the range (1, Infinity) where larger values increase the sparsity of the model

AUXILIARY_3

public static final CSKLR.UpdateMode AUXILIARY_3

Performs model updates based on a "auxiliary" function a(z) = max(loss(z), loss(γ). gamma should be in the range (0, Infinity) where smaller values increase the sparsity of the model

Method Detail

values

public static CSKLR.UpdateMode[] values()

Returns an array containing the constants of this enum type, in the order they are declared. This method may be used to iterate over the constants as follows:

for (CSKLR.UpdateMode c : CSKLR.UpdateMode.values())
    System.out.println(c);

Returns:

an array containing the constants of this enum type, in the order they are declared

valueOf

public static CSKLR.UpdateMode valueOf(String name)

Returns the enum constant of this type with the specified name. The string must match exactly an identifier used to declare an enum constant in this type. (Extraneous whitespace characters are not permitted.)

Parameters:

name - the name of the enum constant to be returned.

Returns:

the enum constant with the specified name

Throws:

IllegalArgumentException - if this enum type has no constant with the specified name

NullPointerException - if the argument is null

pt

protected abstract double pt(double y,
                             double score,
                             double preScore,
                             double eta,
                             double gamma)

Returns the Bernoulli trial probability variable

Parameters:

y - the sign of the input point

score - the logistic regression score for the input

preScore - the raw margin before the final

eta - the learning rate

gamma - the gamma variable

Returns:

the Bernoulli trial probability variable

grad

protected abstract double grad(double y,
                               double score,
                               double preScore,
                               double gamma)

Get the gradient value that should be applied based on the input variable from the current model

Parameters:

y - the sign of the input point

score - the logistic regression score for the input

preScore - the raw margin before the final

gamma - the gamma variable

Returns:

the coefficient to apply to the stochastic update