jsat.math.optimization.stochastic

Class AdaDelta

java.lang.Object

jsat.math.optimization.stochastic.AdaDelta

All Implemented Interfaces:

java.io.Serializable, GradientUpdater

public class AdaDelta
extends java.lang.Object
implements GradientUpdater

AdaDelta is inspired by AdaGrad and was developed for use primarily in neural networks. It still maintains a per feature learning rate, however unlike AdaGrad the learning rates may increase over time and are highly robust to any individual learning rate.

See: Zeiler, M. D. (2012). ADADELTA: An Adaptive Learning Rate Method. CoRR, abs/1212.5.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
AdaDelta() Creates a new AdaDelta updater using a decay rate of 0.95
AdaDelta(AdaDelta toCopy) Copy constructor
AdaDelta(double rho) Creates a new AdaDelta updater

Method Summary

Modifier and Type	Method and Description
AdaDelta	clone()
double	getRho()
void	setRho(double rho) Sets the decay rate used by AdaDelta.
void	setup(int d) Sets up this updater to update a weight vector of dimension d by a gradient of the same dimension
void	update(Vec x, Vec grad, double eta) Updates the weight vector x such that x = x-ηf(grad), where f(grad) is some function on the gradient that effectively returns a new vector.
double	update(Vec x, Vec grad, double eta, double bias, double biasGrad) Updates the weight vector x such that x = x-ηf(grad), where f(grad) is some function on the gradient that effectively returns a new vector.

Methods inherited from class java.lang.Object

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

Constructor Detail

AdaDelta

public AdaDelta()

Creates a new AdaDelta updater using a decay rate of 0.95

AdaDelta

public AdaDelta(double rho)

Creates a new AdaDelta updater

Parameters:

rho - the decay rate to use

AdaDelta

public AdaDelta(AdaDelta toCopy)

Copy constructor

Parameters:

toCopy - the object to copy

Method Detail

setRho

public void setRho(double rho)

Sets the decay rate used by AdaDelta. Lower values focus more on the current gradient, where higher values incorporate a longer history.

Parameters:

rho - the decay rate in (0, 1) to use

getRho

public double getRho()

Returns:

the decay rate that will be used

update

public void update(Vec x,
                   Vec grad,
                   double eta)

Description copied from interface: GradientUpdater

Updates the weight vector x such that x = x-ηf(grad), where f(grad) is some function on the gradient that effectively returns a new vector. It is not necessary for the internal implementation to ever explicitly form any of these objects, so long as x is mutated to have the correct result.

Specified by:

update in interface GradientUpdater

Parameters:

x - the vector to mutate such that is has been updated by the gradient

grad - the gradient to update the weight vector x from

eta - the learning rate to apply

update

public double update(Vec x,
                     Vec grad,
                     double eta,
                     double bias,
                     double biasGrad)

Description copied from interface: GradientUpdater

Updates the weight vector x such that x = x-ηf(grad), where f(grad) is some function on the gradient that effectively returns a new vector. It is not necessary for the internal implementation to ever explicitly form any of these objects, so long as x is mutated to have the correct result.

This version of the update method includes two extra parameters to make it easer to use when a scalar bias term is also used

Specified by:

update in interface GradientUpdater

Parameters:

x - the vector to mutate such that is has been updated by the gradient

grad - the gradient to update the weight vector x from

eta - the learning rate to apply

bias - the bias term of the vector

biasGrad - the gradient for the bias term

Returns:

the value to change the bias by, the update being bias = bias - returnValue

clone

public AdaDelta clone()

Specified by:

clone in interface GradientUpdater

Overrides:

clone in class java.lang.Object

setup

public void setup(int d)

Description copied from interface: GradientUpdater

Sets up this updater to update a weight vector of dimension d by a gradient of the same dimension

Specified by:

setup in interface GradientUpdater

Parameters:

d - the dimension of the weight vector that will be updated