org.tribuo.math.optimisers

Class AdaDelta

java.lang.Object

org.tribuo.math.optimisers.AdaDelta

All Implemented Interfaces:

com.oracle.labs.mlrg.olcut.config.Configurable, com.oracle.labs.mlrg.olcut.provenance.Provenancable<com.oracle.labs.mlrg.olcut.provenance.ConfiguredObjectProvenance>, StochasticGradientOptimiser

public class AdaDelta
extends Object
implements StochasticGradientOptimiser

An implementation of the AdaDelta gradient optimiser.

Creates two copies of the parameters to store learning rates.

See:

 Zeiler, MD.
 "ADADELTA: an Adaptive Learning Rate Method"
 arXiv preprint arXiv:1212.5701.
 

Constructor Summary

Constructors

Constructor and Description
AdaDelta() Sets rho to 0.95 and epsilon to 1e-6.
AdaDelta(double epsilon) Keeps rho at 0.95, passes through epsilon.
AdaDelta(double rho, double epsilon) It's recommended to keep rho at 0.95.

Method Summary

Modifier and Type	Method and Description
AdaDelta	copy() Copies a gradient optimiser with it's configuration.
com.oracle.labs.mlrg.olcut.provenance.ConfiguredObjectProvenance	getProvenance()
void	initialise(Parameters parameters) Initialises the gradient optimiser.
void	reset() Resets the optimiser so it's ready to optimise a new Parameters.
Tensor[]	step(Tensor[] updates, double weight) Take a Tensor array of gradients and transform them according to the current weight and learning rates.
String	toString()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface org.tribuo.math.StochasticGradientOptimiser

finalise

Methods inherited from interface com.oracle.labs.mlrg.olcut.config.Configurable

postConfig

Constructor Detail

AdaDelta

public AdaDelta(double rho,
                double epsilon)

It's recommended to keep rho at 0.95.

Parameters:

rho - The rho value.

epsilon - The epsilon value.

AdaDelta

public AdaDelta(double epsilon)

Keeps rho at 0.95, passes through epsilon.

Parameters:

epsilon - The epsilon value.

AdaDelta

public AdaDelta()

Sets rho to 0.95 and epsilon to 1e-6.

Method Detail

initialise

public void initialise(Parameters parameters)

Description copied from interface: StochasticGradientOptimiser

Initialises the gradient optimiser.

Configures any learning rate parameters.

Specified by:

initialise in interface StochasticGradientOptimiser

Parameters:

parameters - The parameters to optimise.

step

public Tensor[] step(Tensor[] updates,
                     double weight)

Description copied from interface: StochasticGradientOptimiser

Take a Tensor array of gradients and transform them according to the current weight and learning rates.

Can return the same Tensor array or a new one.

Specified by:

step in interface StochasticGradientOptimiser

Parameters:

updates - An array of gradients.

weight - The weight for the current gradients.

Returns:

A Tensor array of gradients.

toString

public String toString()

Overrides:

toString in class Object

reset

public void reset()

Description copied from interface: StochasticGradientOptimiser

Resets the optimiser so it's ready to optimise a new Parameters.

Specified by:

reset in interface StochasticGradientOptimiser

copy

public AdaDelta copy()

Description copied from interface: StochasticGradientOptimiser

Copies a gradient optimiser with it's configuration. Usually calls the copy constructor.

Specified by:

copy in interface StochasticGradientOptimiser

Returns:

A gradient optimiser with the same configuration, but independent state.

getProvenance

public com.oracle.labs.mlrg.olcut.provenance.ConfiguredObjectProvenance getProvenance()

Specified by:

getProvenance in interface com.oracle.labs.mlrg.olcut.provenance.Provenancable<com.oracle.labs.mlrg.olcut.provenance.ConfiguredObjectProvenance>