org.aika.neuron

Class Activation

java.lang.Object

org.aika.lattice.NodeActivation<OrNode>

org.aika.neuron.Activation

All Implemented Interfaces:

Comparable<NodeActivation>

public final class Activation
extends NodeActivation<OrNode>

The Activation class is the most central class in Aika. On the one hand it stores the activation value for a given neuron in the State substructure. On the other hand it specifies where this activation is located within the document and to which interpretation it belongs. The Activation.Key therefore consists of the logic node to which this activation belongs. If this logic node is an or-node, then this activation automatically also belongs to the neuron as well. Furthermore, the key contains the char range within the document and the relational id (rid). The relational id might be used to store the word pos for instance. Lastly, the key contain the interpretation node of this activation, specifying to which interpretation this activation belongs.

The activations are linked to each other on two levels. The fields inputs and outputs contain the activation links within the logic layer. The fields neuronInputs and neuronOutputs contain the links on the neural layer.

Author:

Lukas Molzberger

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	Activation.Builder
static class	Activation.Mode
static class	Activation.Rounds Since Aika is a recurrent neural network, it is necessary to compute several rounds of activation values.
static class	Activation.State A State object contains the activation value of an activation object that belongs to a neuron.
class	Activation.StateChange The StateChange class is used to store the state change of an activation that occurs in each node of the binary search tree.
static class	Activation.SynapseActivation The SynapseActivation mirror the synapse link in the network of activations.

Nested classes/interfaces inherited from class org.aika.lattice.NodeActivation

NodeActivation.Key<T extends Node>

Field Summary

Fields

Modifier and Type	Field and Description
static Comparator<Activation>	ACTIVATION_ID_COMP
Activation.StateChange	currentStateChange
long	currentStateV
double	errorSignal
Double	inputValue
boolean	isQueued
double	lowerBound
long	markedDirty
double	maxActValue
TreeSet<Activation.SynapseActivation>	neuronInputs
TreeSet<Activation.SynapseActivation>	neuronOutputs
long	queueId
Activation.Rounds	rounds
Integer	sequence
Double	targetValue
boolean	ubQueued
double	upperBound

Fields inherited from class org.aika.lattice.NodeActivation

doc, id, inputs, key, outputs, recomputeUBV, repropagateV, visited

Constructor Summary

Constructors

Constructor and Description
Activation(int id, Document doc, NodeActivation.Key key)

Method Summary

Modifier and Type	Method and Description
void	addSynapseActivation(int dir, Activation.SynapseActivation sa)
void	computeBounds()
Activation.State	computeValueAndWeight(int round)
<T extends Node> boolean	filter(T n, Integer rid, Range r, Range.Relation rr, InterpretationNode o, InterpretationNode.Relation or)
static Activation	get(Document doc, INeuron n, Integer rid, Range r, Range.Relation rr, InterpretationNode o, InterpretationNode.Relation or)
static Activation	get(Document doc, INeuron n, NodeActivation.Key ak)
static Stream<Activation>	getActivationsByRange(INeuron.ThreadState th, INeuron n, Integer rid, Range r, Range.Relation rr, InterpretationNode o, InterpretationNode.Relation or)
List<Activation.SynapseActivation>	getFinalInputActivations()
List<Activation.SynapseActivation>	getFinalOutputActivations()
Activation.State	getFinalState()
INeuron	getINeuron()
String	getLabel()
Neuron	getNeuron()
Integer	getSequence()
boolean	isFinalActivation()
String	linksToString()
void	markDirty(long v)
SearchNode.Weight	process(SearchNode sn, int round, long v)
void	processBounds()
void	saveNewState()
void	saveOldState(Map<Activation,Activation.StateChange> changes, long v)
static Stream<Activation>	select(Document doc, INeuron n, Integer rid, Range r, Range.Relation rr, InterpretationNode o, InterpretationNode.Relation or)
static Stream<Activation>	select(INeuron.ThreadState th, INeuron n, Integer rid, Range r, Range.Relation rr, InterpretationNode o, InterpretationNode.Relation or)
void	setTargetValue(Double targetValue)
String	toString()
String	toString(SearchNode sn, boolean withTextSnippet, boolean withLogic)

Methods inherited from class org.aika.lattice.NodeActivation

compare, compareTo, link, toString

Methods inherited from class java.lang.Object

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

Field Detail

ACTIVATION_ID_COMP

public static final Comparator<Activation> ACTIVATION_ID_COMP

neuronInputs

public TreeSet<Activation.SynapseActivation> neuronInputs

neuronOutputs

public TreeSet<Activation.SynapseActivation> neuronOutputs

sequence

public Integer sequence

upperBound

public double upperBound

lowerBound

public double lowerBound

rounds

public Activation.Rounds rounds

maxActValue

public double maxActValue

ubQueued

public boolean ubQueued

isQueued

public boolean isQueued

queueId

public long queueId

currentStateV

public long currentStateV

currentStateChange

public Activation.StateChange currentStateChange

markedDirty

public long markedDirty

errorSignal

public double errorSignal

targetValue

public Double targetValue

inputValue

public Double inputValue

Constructor Detail

Activation

public Activation(int id,
                  Document doc,
                  NodeActivation.Key key)

Method Detail

setTargetValue

public void setTargetValue(Double targetValue)

getLabel

public String getLabel()

getINeuron

public INeuron getINeuron()

getNeuron

public Neuron getNeuron()

addSynapseActivation

public void addSynapseActivation(int dir,
                                 Activation.SynapseActivation sa)

process

public SearchNode.Weight process(SearchNode sn,
                                 int round,
                                 long v)

computeValueAndWeight

public Activation.State computeValueAndWeight(int round)

processBounds

public void processBounds()

computeBounds

public void computeBounds()

getFinalInputActivations

public List<Activation.SynapseActivation> getFinalInputActivations()

getFinalOutputActivations

public List<Activation.SynapseActivation> getFinalOutputActivations()

isFinalActivation

public boolean isFinalActivation()

getFinalState

public Activation.State getFinalState()

get

public static Activation get(Document doc,
                             INeuron n,
                             Integer rid,
                             Range r,
                             Range.Relation rr,
                             InterpretationNode o,
                             InterpretationNode.Relation or)

get

public static Activation get(Document doc,
                             INeuron n,
                             NodeActivation.Key ak)

select

public static Stream<Activation> select(Document doc,
                                        INeuron n,
                                        Integer rid,
                                        Range r,
                                        Range.Relation rr,
                                        InterpretationNode o,
                                        InterpretationNode.Relation or)

select

public static Stream<Activation> select(INeuron.ThreadState th,
                                        INeuron n,
                                        Integer rid,
                                        Range r,
                                        Range.Relation rr,
                                        InterpretationNode o,
                                        InterpretationNode.Relation or)

getActivationsByRange

public static Stream<Activation> getActivationsByRange(INeuron.ThreadState th,
                                                       INeuron n,
                                                       Integer rid,
                                                       Range r,
                                                       Range.Relation rr,
                                                       InterpretationNode o,
                                                       InterpretationNode.Relation or)

filter

public <T extends Node> boolean filter(T n,
                                       Integer rid,
                                       Range r,
                                       Range.Relation rr,
                                       InterpretationNode o,
                                       InterpretationNode.Relation or)

getSequence

public Integer getSequence()

markDirty

public void markDirty(long v)

toString

public String toString()

Overrides:

toString in class Object

toString

public String toString(SearchNode sn,
                       boolean withTextSnippet,
                       boolean withLogic)

linksToString

public String linksToString()

saveOldState

public void saveOldState(Map<Activation,Activation.StateChange> changes,
                         long v)

saveNewState

public void saveNewState()