org.aika

Class Model

java.lang.Object

org.aika.Model

public class Model
extends Object

The model consists of two layers. The first layer is the actual neural network consisting of neurons and synapses. The second layer is a pattern lattice containing a boolean logic representation of all the neurons. Whenever the synapse weights of a neuron are adjusted, then the underlying boolean logic representation of this neuron will be updated too.

The model supports parallel processing using a fixed number of threads.

Author:

Lukas Molzberger

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	Model.Statistic

Field Summary

Fields

Modifier and Type	Field and Description
Map<Integer,Provider<? extends AbstractNode>>	activeProviders
AtomicInteger	currentId
int	defaultThreadId
Document[]	docs
int[]	lastCleanup
int	numberOfPositions
Set<Provider<AndNode>>	numberOfPositionsQueue
int	numberOfThreads
WeakHashMap<Integer,WeakReference<Provider<? extends AbstractNode>>>	providers
Model.Statistic	stat
SuspensionHook	suspensionHook

Constructor Summary

Constructors

Constructor and Description
Model() Creates a model with a single thread.
Model(SuspensionHook sh, int numberOfThreads)

Method Summary

Modifier and Type	Method and Description
void	addSynapse(Neuron n, Input input)
Document	createDocument(String txt)
Document	createDocument(String txt, int threadId)
Neuron	createNeuron()
Neuron	createNeuron(String label)
Neuron	createNeuron(String label, boolean isBlocked, boolean noTraining)
void	discardAll() Discards all unsuspended neurons and logic nodes.
Neuron	initCounterNeuron(Neuron n, Neuron clockSignal, boolean dirCS, Neuron startSignal, boolean dirSS, boolean direction) A counter neuron is used to compute the relational id.
Neuron	initNeuron(Neuron n, double bias, Collection<Input> inputs) Creates a neuron with the given bias.
Neuron	initNeuron(Neuron n, double bias, Input... inputs) Creates a neuron with the given bias.
Neuron	initRelationalNeuron(Neuron n, Neuron ctn, Neuron inputSignal, boolean dirIS) A relational neuron combines the relational id created by a counter neuron with an input signal.
Neuron	lookupNeuron(int id)
<P extends Provider<? extends Node>> P	lookupNodeProvider(int id)
void	register(Provider p)
void	resetFrequency()
void	suspendAll() Suspend all neurons and logic nodes in memory.
void	suspendUnusedNodes(int docId) Suspend all neurons and logic nodes whose last used document id is lower/older than .
void	unregister(Provider p)

Methods inherited from class java.lang.Object

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

Field Detail

numberOfThreads

public int numberOfThreads

lastCleanup

public int[] lastCleanup

docs

public Document[] docs

suspensionHook

public SuspensionHook suspensionHook

currentId

public AtomicInteger currentId

providers

public WeakHashMap<Integer,WeakReference<Provider<? extends AbstractNode>>> providers

activeProviders

public Map<Integer,Provider<? extends AbstractNode>> activeProviders

stat

public Model.Statistic stat

defaultThreadId

public int defaultThreadId

numberOfPositionsQueue

public Set<Provider<AndNode>> numberOfPositionsQueue

numberOfPositions

public volatile int numberOfPositions

Constructor Detail

Model

public Model()

Creates a model with a single thread.

Model

public Model(SuspensionHook sh,
             int numberOfThreads)

Method Detail

createNeuron

public Neuron createNeuron()

createNeuron

public Neuron createNeuron(String label)

createNeuron

public Neuron createNeuron(String label,
                           boolean isBlocked,
                           boolean noTraining)

createDocument

public Document createDocument(String txt)

createDocument

public Document createDocument(String txt,
                               int threadId)

lookupNodeProvider

public <P extends Provider<? extends Node>> P lookupNodeProvider(int id)

lookupNeuron

public Neuron lookupNeuron(int id)

resetFrequency

public void resetFrequency()

register

public void register(Provider p)

unregister

public void unregister(Provider p)

suspendUnusedNodes

public void suspendUnusedNodes(int docId)

Suspend all neurons and logic nodes whose last used document id is lower/older than .

Parameters:

docId -

suspendAll

public void suspendAll()

Suspend all neurons and logic nodes in memory.

discardAll

public void discardAll()

Discards all unsuspended neurons and logic nodes.

initNeuron

public Neuron initNeuron(Neuron n,
                         double bias,
                         Collection<Input> inputs)

Creates a neuron with the given bias.

Parameters:

n -

bias -

inputs -

Returns:

addSynapse

public void addSynapse(Neuron n,
                       Input input)

initNeuron

public Neuron initNeuron(Neuron n,
                         double bias,
                         Input... inputs)

Creates a neuron with the given bias.

Parameters:

n -

bias -

inputs -

Returns:

initRelationalNeuron

public Neuron initRelationalNeuron(Neuron n,
                                   Neuron ctn,
                                   Neuron inputSignal,
                                   boolean dirIS)

A relational neuron combines the relational id created by a counter neuron with an input signal.

Parameters:

n -

ctn -

inputSignal -

dirIS -

Returns:

initCounterNeuron

public Neuron initCounterNeuron(Neuron n,
                                Neuron clockSignal,
                                boolean dirCS,
                                Neuron startSignal,
                                boolean dirSS,
                                boolean direction)

A counter neuron is used to compute the relational id. It simply adds a new activation after each clock signal and increases the relational id by one.

Parameters:

n -

clockSignal -

dirCS - The direction of the clock signal.

startSignal -

dirSS - The direction of the start signal.

direction -

Returns: