org.joone.engine
Class RTRLLearnerFactory.Weight

java.lang.Object
  org.joone.engine.RTRLLearnerFactory.Weight

All Implemented Interfaces:

java.io.Serializable

Enclosing class:

RTRLLearnerFactory

protected class RTRLLearnerFactory.Weight
extends java.lang.Object
implements java.io.Serializable

A weight. In order to populate the weight matrix, we need to know where to find the weight in joone. The weight can either be a bias found inside a layer or a normal weight found inside a synapse. This class is used to store the weight as well as the indices needed to reference it inside the joone object. We also store a lot of indexing information, such as k and u, in this class for performance reasons. We try to iterate weights rather than loop some index i from 0 to n - 1 for example. This also applies elsewhere in this implementation.

See Also:

Serialized Form

Field Summary

protected int	i The joone i index of this weight
protected int	j The joone j index of this weight
protected int	K The node K in Z from which this weight comes
protected Layer	layer The joone layer which is used if this weight is a bias
protected Synapse	synapse The joone synapse which is used if this weight is a normal synaptical weight

Constructor Summary

RTRLLearnerFactory.Weight(Layer layer, int i, int K)
Initialise this weight from a joone layer

RTRLLearnerFactory.Weight(Synapse synapse, int i, int j, int K)
Initialise this weight from a joone synapse.

Method Summary

void	addDelta(double delta) Add a given amount to this weight's delta
void	addWeight(double delta) Update this weight with the given delta amount, also update the weight delta
double	getDelta() Retrieve this weight's delta
double	getWeight() Retrieve this weight
void	setDelta(double delta) Set this weight's delta to a specific value
void	setWeight(double weight) Set this weight to a specific value

Methods inherited from class java.lang.Object

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

Field Detail

K

protected int K

The node K in Z from which this weight comes

i

protected int i

The joone i index of this weight

j

protected int j

The joone j index of this weight

layer

protected Layer layer

The joone layer which is used if this weight is a bias

synapse

protected Synapse synapse

The joone synapse which is used if this weight is a normal synaptical weight

Constructor Detail

RTRLLearnerFactory.Weight

public RTRLLearnerFactory.Weight(Layer layer,
                                 int i,
                                 int K)

Initialise this weight from a joone layer

Parameters:

layer - the joone layer where this is found

i - the index into the layer at which this is found

K - the index

RTRLLearnerFactory.Weight

public RTRLLearnerFactory.Weight(Synapse synapse,
                                 int i,
                                 int j,
                                 int K)

Initialise this weight from a joone synapse. Note that joone stores its weights in columns and rows, not the usual rows and columns. So we swap i and j in this constructor.

Parameters:

synapse - the joone synapse where this is found

i - the row in the synapse at which the weight is found

j - the column in the synapse at which the weight is found

K - the node in Z from which this weight is coming

Method Detail

addWeight

public void addWeight(double delta)

Update this weight with the given delta amount, also update the weight delta

setWeight

public void setWeight(double weight)

Set this weight to a specific value

getWeight

public double getWeight()

Retrieve this weight

addDelta

public void addDelta(double delta)

Add a given amount to this weight's delta

setDelta

public void setDelta(double delta)

Set this weight's delta to a specific value

getDelta

public double getDelta()

Retrieve this weight's delta