uchicago.src.sim.adaptation.neural
Class NeuralUtils

java.lang.Object
  uchicago.src.sim.adaptation.neural.NeuralUtils

public class NeuralUtils
extends java.lang.Object

Some utility functions for creating neural networks.

Version:

$Revision: 1.3 $ $Date: 2005/08/12 16:13:29 $

Author:

Jerry Vos

See Also:

RepastNeuralWrapper

Method Summary

static RepastNeuralWrapper	buildNetwork(int[] nodesPerLayer, java.lang.Class[] layerTypes, java.lang.Class[] synapseTypes) This builds a neural network using the specified parameters.
static RepastNeuralWrapper	buildNetwork(int[] nodesPerLayer, java.lang.Class layerType, java.lang.Class synapseType) This builds a neural network with nodesPerLayer.length layers, each layer being an instance of layerType, and each layer connected by a synapse of type synapseType.
static org.joone.engine.Synapse	linkLayers(org.joone.engine.Layer outLayer, org.joone.engine.Layer inLayer, java.lang.Class synapseType) Links two layers using the specified synapse type.
static RepastNeuralWrapper	loadNetFromFile(java.lang.String fileName) This method creates a RepastNeuralWrapper that is wrapping a serialized Joone NeuralNet.
static void	saveNetToFile(org.joone.net.NeuralNet net, java.lang.String fileName) Saves a NeuralNet to the specified file.

Methods inherited from class java.lang.Object

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

Method Detail

buildNetwork

public static RepastNeuralWrapper buildNetwork(int[] nodesPerLayer,
                                               java.lang.Class layerType,
                                               java.lang.Class synapseType)
                                        throws NeuralException

This builds a neural network with nodesPerLayer.length layers, each layer being an instance of layerType, and each layer connected by a synapse of type synapseType.

Parameters:

nodesPerLayer - the number of nodes in the layers

layerType - the type to make the layers

synapseType - the type to make the synapses

Returns:

the created network

Throws:

NeuralException - when there is an error creating the network

See Also:

buildNetwork(int[], Class[], Class[])

buildNetwork

public static RepastNeuralWrapper buildNetwork(int[] nodesPerLayer,
                                               java.lang.Class[] layerTypes,
                                               java.lang.Class[] synapseTypes)
                                        throws NeuralException,
                                               java.lang.IllegalArgumentException

This builds a neural network using the specified parameters. This network will have all the layers connected using the synapse types passed in. No input or output synapses will be attached to the layer, so you still must do this yourself.
This network will also have a teacher added to it, however, this teacher will NOT be attached as an output layer; it only will be added to the network.

Parameters:

nodesPerLayer - the number of nodes in each layer

layerTypes - the types to make the layers

synapseTypes - the types of the synapses connecting the layers

Returns:

the created network

Throws:

NeuralException - when there is an error instantiating the layers/synapses

java.lang.IllegalArgumentException - when (nodesPerLayer.length != layerTypes.length || layerTypes.length -1 != synapseTypes.length)

linkLayers

public static org.joone.engine.Synapse linkLayers(org.joone.engine.Layer outLayer,
                                                  org.joone.engine.Layer inLayer,
                                                  java.lang.Class synapseType)
                                           throws NeuralException

Links two layers using the specified synapse type.

Parameters:

outLayer - the layer the synapse is coming out of

inLayer - the layer the synapse is going into

synapseType - the type of the synapse

Returns:

the created synapse instance

Throws:

NeuralException - when there is an error instantiating the synapse

saveNetToFile

public static void saveNetToFile(org.joone.net.NeuralNet net,
                                 java.lang.String fileName)
                          throws NeuralException

Saves a NeuralNet to the specified file.

Parameters:

net - the neural network to save

fileName - the name of the file to save the network to

Throws:

NeuralException - when there is an error saving the network

See Also:

RepastNeuralWrapper.saveNetToFile(String), RepastNeuralWrapper#loadNetFromFile(String)

loadNetFromFile

public static RepastNeuralWrapper loadNetFromFile(java.lang.String fileName)
                                           throws NeuralException

This method creates a RepastNeuralWrapper that is wrapping a serialized Joone NeuralNet. In other words, this creates a wrapper around a predesigned NeuralNet.

Parameters:

fileName - the name of the serialized NeuralNet

Returns:

the created wrapper

Throws:

NeuralException - when there is an error loading the specified network