de.up.ling.irtg.learning_rates

Class AdaGrad

java.lang.Object

de.up.ling.irtg.learning_rates.AdaGrad

All Implemented Interfaces:

LearningRate

public class AdaGrad
extends Object
implements LearningRate

A simple implementation of the widely used AdaGrad learning rate.

Author:

teichmann

See Also:

Duchi and Elad Hazan and Yoram Singer, Adaptive Subgradient Methods for Online Learning and Stochastic Optimization, 2010, University of California, Berkeley

Constructor Summary

Constructors

Constructor and Description
AdaGrad() Creates and instance with a base rate of 0.5.
AdaGrad(double baseRate) Create a new instance with the given base rate.

Method Summary

Modifier and Type	Method and Description
double	getLearningRate(int group, int number, double gradient) Returns the learning rate for the parameter in the given group with the given number.
void	reset() Resets the learning rate sequence to its initial state.

Methods inherited from class java.lang.Object

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

Constructor Detail

AdaGrad

public AdaGrad(double baseRate)

Create a new instance with the given base rate.

Parameters:

baseRate -

AdaGrad

public AdaGrad()

Creates and instance with a base rate of 0.5.

Method Detail

getLearningRate

public double getLearningRate(int group,
                              int number,
                              double gradient)

Description copied from interface: LearningRate

Returns the learning rate for the parameter in the given group with the given number. The gradient passed should be the gradient in the current iterate.

Specified by:

getLearningRate in interface LearningRate

Returns:

reset

public void reset()

Description copied from interface: LearningRate

Resets the learning rate sequence to its initial state.

Specified by:

reset in interface LearningRate