at.tuwien.ifs.somtoolbox.models

Class GHSOM

java.lang.Object

at.tuwien.ifs.somtoolbox.models.AbstractNetworkModel

at.tuwien.ifs.somtoolbox.models.GHSOM

All Implemented Interfaces:

SOMToolboxApp, NetworkModel

public class GHSOM
extends AbstractNetworkModel
implements SOMToolboxApp

This class implements the Growing Hierarchical Self-Organizing Map. It is basically a wrapper for the GrowingSOM and mainly handles command line execution and parameters. It implements the NetworkModel interface which is currently not used, but may be used in the future. It is also not clear, if this class will be removed and replaced by the GrowingSOM, because it already contains the hierarchical functionality, only the training procedure would have to be updated.

Version:

$Id: GHSOM.java 3995 2011-01-19 14:50:20Z mayer $

Author:

Michael Dittenbach

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
private class	GHSOM.ExpandedUnits

Nested classes/interfaces inherited from interface at.tuwien.ifs.somtoolbox.apps.SOMToolboxApp

SOMToolboxApp.Type

Field Summary

Fields

Modifier and Type	Field and Description
static SOMToolboxApp.Type	APPLICATION_TYPE
static java.lang.String	DESCRIPTION
private GrowingLayer	layer0
static java.lang.String	LONG_DESCRIPTION
static com.martiansoftware.jsap.Parameter[]	OPTIONS
private GrowingSOM	topLayerMap

Fields inherited from class at.tuwien.ifs.somtoolbox.models.AbstractNetworkModel

contentType, DEFAULT_LABEL_COUNT, labelled, sharedInputObjects, trainingStart

Fields inherited from interface at.tuwien.ifs.somtoolbox.apps.SOMToolboxApp

DEV_BY_STRING, HOMEPAGE, INFO, INFO_TEXT, LICENSE_TEXT, REQUIRED_MEMBERS, TYPE_GROUPED_COMPARATOR

Constructor Summary

Constructors

Constructor and Description
GHSOM(int dim, boolean norm, GHSOMProperties props, InputData data) Constructs a new GHSOM with dim-dimensional weight vectors.
GHSOM(SOMInputReader ir) Constructs an already trained GHSOM with a SOMInputReader provided by argument ir.

Method Summary

Modifier and Type	Method and Description
GHSOM.ExpandedUnits	getExpandedUnits(GrowingLayer layer, QualityMeasure qm, java.lang.String qmName, double fraction, double totalQuality)
static void	main(java.lang.String[] args) Method for stand-alone execution of map training. Options are: -h toggles HTML output -l name of class implementing the labeling algorithm -n number of labels to generate -w name of weight vector file in case of training an already trained map -m name of map description file in case of training an already trained map --noDWM switch to not write the data winner mapping file properties name of properties file, mandatory
void	setSharedInputObjects(SharedSOMVisualisationData sharedInputObjects)
GrowingSOM	topLayerMap() Returns the top-layer map
void	train(InputData data, GHSOMProperties props) Trains the GHSOM with the input data and training parameters specified in the properties provided by argument props.

Methods inherited from class at.tuwien.ifs.somtoolbox.models.AbstractNetworkModel

getDataContentType, getInputData, getSharedInputObjects, isLabelled, printTrainingTime, setLabelled

Methods inherited from class java.lang.Object

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

Field Detail

DESCRIPTION

public static java.lang.String DESCRIPTION

APPLICATION_TYPE

public static final SOMToolboxApp.Type APPLICATION_TYPE

LONG_DESCRIPTION

public static java.lang.String LONG_DESCRIPTION

OPTIONS

public static final com.martiansoftware.jsap.Parameter[] OPTIONS

layer0

private GrowingLayer layer0

topLayerMap

private GrowingSOM topLayerMap

Constructor Detail

GHSOM

public GHSOM(int dim,
             boolean norm,
             GHSOMProperties props,
             InputData data)

Constructs a new GHSOM with dim-dimensional weight vectors. Argument norm determines whether the randomly initialised weight vectors should be normalised to unit length or not.

Parameters:

dim - the dimensionality of the weight vectors.

norm - specifies if the weight vectors are to be normalised to unit length.

props - the network properties.

GHSOM

public GHSOM(SOMInputReader ir)

Constructs an already trained GHSOM with a SOMInputReader provided by argument ir.

Parameters:

ir - an object implementing the SOMinputReader interface to load an already trained model.

Method Detail

main

public static void main(java.lang.String[] args)

Method for stand-alone execution of map training.
Options are:

• -h toggles HTML output
• -l name of class implementing the labeling algorithm
• -n number of labels to generate
• -w name of weight vector file in case of training an already trained map
• -m name of map description file in case of training an already trained map
• --noDWM switch to not write the data winner mapping file
• properties name of properties file, mandatory

Parameters:

args - the execution arguments as stated above.

setSharedInputObjects

public void setSharedInputObjects(SharedSOMVisualisationData sharedInputObjects)

Overrides:

setSharedInputObjects in class AbstractNetworkModel

getExpandedUnits

public GHSOM.ExpandedUnits getExpandedUnits(GrowingLayer layer,
                                            QualityMeasure qm,
                                            java.lang.String qmName,
                                            double fraction,
                                            double totalQuality)

topLayerMap

public GrowingSOM topLayerMap()

Returns the top-layer map

Returns:

the top-layer map.

train

public void train(InputData data,
                  GHSOMProperties props)

Trains the GHSOM with the input data and training parameters specified in the properties provided by argument props.

Parameters:

data - input data to train the map with.

props - the training properties.