org.encog.neural.networks.training.som
Class TrainSelfOrganizingMap

java.lang.Object
  org.encog.neural.networks.training.som.TrainSelfOrganizingMap

All Implemented Interfaces:

Train

public class TrainSelfOrganizingMap
extends java.lang.Object
implements Train

TrainSelfOrganizingMap: Implements an unsupervised training algorithm for use with a Self Organizing Map.

Nested Class Summary

static class

TrainSelfOrganizingMap.LearningMethod The learning method, either additive or subtractive.

Field Summary

static double	DEFAULT_REDUCTION The default reduction to use.
static double	MIN_LEARNRATE_FOR_REDUCTION The minimum learning rate for reduction to be applied.

Constructor Summary

TrainSelfOrganizingMap(BasicNetwork network, NeuralDataSet train, TrainSelfOrganizingMap.LearningMethod learnMethod, double learnRate)
Construct the trainer for a self organizing map.

Method Summary

protected void	adjustWeights() Adjust the weights and allow the network to learn.
void	evaluateErrors() Evaludate the current error level of the network.
protected void	forceWin() Force a win, if no neuron won.
double	getBestError() Get the best error so far.
double	getError() Get the current error percent from the training.
Network	getNetwork() Get the current best network from the training.
double	getTotalError() Get the error for this iteration.
void	initialize() Called to initialize the SOM.
void	iteration() This method is called for each training iteration.
protected void	normalizeWeight(Matrix matrix, int row) Normalize the specified row in the weight matrix.

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_REDUCTION

public static final double DEFAULT_REDUCTION

The default reduction to use.

See Also:

Constant Field Values

MIN_LEARNRATE_FOR_REDUCTION

public static final double MIN_LEARNRATE_FOR_REDUCTION

The minimum learning rate for reduction to be applied.

See Also:

Constant Field Values

Constructor Detail

TrainSelfOrganizingMap

public TrainSelfOrganizingMap(BasicNetwork network,
                              NeuralDataSet train,
                              TrainSelfOrganizingMap.LearningMethod learnMethod,
                              double learnRate)

Construct the trainer for a self organizing map.

Parameters:

network - The network to train.

train - The training method.

learnMethod - The learning method.

learnRate - The learning rate.

Method Detail

adjustWeights

protected void adjustWeights()

Adjust the weights and allow the network to learn.

evaluateErrors

public void evaluateErrors()

Evaludate the current error level of the network.

forceWin

protected void forceWin()

Force a win, if no neuron won.

getBestError

public double getBestError()

Get the best error so far.

Returns:

The best error so far.

getTotalError

public double getTotalError()

Get the error for this iteration.

Returns:

The error for this iteration.

initialize

public void initialize()

Called to initialize the SOM.

iteration

public void iteration()

This method is called for each training iteration. Usually this method is called from inside a loop until the error level is acceptable.

Specified by:

iteration in interface Train

normalizeWeight

protected void normalizeWeight(Matrix matrix,
                               int row)

Normalize the specified row in the weight matrix.

Parameters:

matrix - The weight matrix.

row - The row to normalize.

getError

public double getError()

Description copied from interface: Train

Get the current error percent from the training.

Specified by:

getError in interface Train

Returns:

The error from the last iteration.

getNetwork

public Network getNetwork()

Description copied from interface: Train

Get the current best network from the training.

Specified by:

getNetwork in interface Train

Returns:

The trained network.