nuroko.module

Class CompoundLayerStack

java.lang.Object

nuroko.module.AComponent

nuroko.module.ALayerStack

nuroko.module.CompoundLayerStack

All Implemented Interfaces:

Cloneable, IComponent, IGradientTrainable, IInput, IInputOutput, IInputState, IModule, IOutput, IOutputState, IParameterised, IThinker, ITrainable

public class CompoundLayerStack
extends ALayerStack

Constructor Summary

Constructors

Constructor and Description
CompoundLayerStack(ALayerStack a, ALayerStack b)

Method Summary

Methods

Modifier and Type	Method and Description
ALayerStack	clone() Creates a clone of a module, including a deep copy of any mutable state.
static ALayerStack	create(List<AWeightLayer> layers)
List<IComponent>	getComponents()
mikera.vectorz.AVector	getData(int i)
mikera.vectorz.AVector	getGradient() Return an AVector referencing the accumulated gradient in this model
mikera.vectorz.AVector	getInputGradient()
AWeightLayer	getLayer(int i)
int	getLayerCount()
mikera.vectorz.AVector	getOutputGradient()
mikera.vectorz.AVector	getParameters() Return an AVector referring to the parameters in the model.
static ALayerStack	stack(ALayerStack a, ALayerStack b)
void	thinkInternal() Thinks within the scope of the component.
void	train(mikera.vectorz.AVector input, mikera.vectorz.AVector target)
void	trainGradient(mikera.vectorz.AVector input, mikera.vectorz.AVector outputGradient, mikera.vectorz.AVector inputGradient, double factor, boolean skipTopDerivative) Trains with a direct gradient.
void	trainGradient(mikera.vectorz.AVector gradient, double factor) Trains with a output gradient, incrementing inputGradient and accumulated gradient for parameters.
void	trainGradientInternal(double factor)

Methods inherited from class nuroko.module.ALayerStack

getInput, getLayers, getOutput, subStack

Methods inherited from class nuroko.module.AComponent

applyConstraints, generate, generate, getComponent, getInputLength, getInputState, getModules, getOutputLength, getParameterLength, isStochastic, setInput, setOutput, think, think, thinkInternalTraining, topComponent, train

Methods inherited from class java.lang.Object

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

Constructor Detail

CompoundLayerStack

public CompoundLayerStack(ALayerStack a,
                  ALayerStack b)

Method Detail

stack

public static ALayerStack stack(ALayerStack a,
                ALayerStack b)

thinkInternal

public void thinkInternal()

Description copied from interface: IComponent

Thinks within the scope of the component. Updates component's output.

trainGradientInternal

public void trainGradientInternal(double factor)

Specified by:

trainGradientInternal in interface IComponent

Specified by:

trainGradientInternal in class AComponent

getComponents

public List<IComponent> getComponents()

getParameters

public mikera.vectorz.AVector getParameters()

Description copied from interface: IParameterised

Return an AVector referring to the parameters in the model. Expected to be mutable.

Returns:

getGradient

public mikera.vectorz.AVector getGradient()

Description copied from interface: IParameterised

Return an AVector referencing the accumulated gradient in this model

Returns:

trainGradient

public void trainGradient(mikera.vectorz.AVector gradient,
                 double factor)

Description copied from interface: IGradientTrainable

Trains with a output gradient, incrementing inputGradient and accumulated gradient for parameters.

Specified by:

trainGradient in interface IGradientTrainable

Overrides:

trainGradient in class AComponent

factor - TODO

getInputGradient

public mikera.vectorz.AVector getInputGradient()

getOutputGradient

public mikera.vectorz.AVector getOutputGradient()

getLayerCount

public int getLayerCount()

Specified by:

getLayerCount in class ALayerStack

getLayer

public AWeightLayer getLayer(int i)

Specified by:

getLayer in class ALayerStack

getData

public mikera.vectorz.AVector getData(int i)

Specified by:

getData in class ALayerStack

clone

public ALayerStack clone()

Description copied from interface: IModule

Creates a clone of a module, including a deep copy of any mutable state.

Specified by:

clone in interface IComponent

Specified by:

clone in interface IModule

Specified by:

clone in interface IParameterised

Specified by:

clone in interface IThinker

Specified by:

clone in interface ITrainable

Specified by:

clone in class ALayerStack

Returns:

train

public void train(mikera.vectorz.AVector input,
         mikera.vectorz.AVector target)

Specified by:

train in interface ITrainable

Overrides:

train in class AComponent

trainGradient

public void trainGradient(mikera.vectorz.AVector input,
                 mikera.vectorz.AVector outputGradient,
                 mikera.vectorz.AVector inputGradient,
                 double factor,
                 boolean skipTopDerivative)

Description copied from class: ALayerStack

Trains with a direct gradient. Assumes think has been called immediately prior to set any intermediate values.

Specified by:

trainGradient in class ALayerStack

create

public static ALayerStack create(List<AWeightLayer> layers)