org.apache.commons.math3.ml.neuralnet.twod.util

Class UnifiedDistanceMatrix

java.lang.Object

org.apache.commons.math3.ml.neuralnet.twod.util.UnifiedDistanceMatrix

All Implemented Interfaces:

MapVisualization

public class UnifiedDistanceMatrix
extends java.lang.Object
implements MapVisualization

U-Matrix visualization of high-dimensional data projection.

Since:

3.6

Field Summary

Fields

Modifier and Type	Field and Description
private DistanceMeasure	distance Distance.
private boolean	individualDistances Whether to show distance between each pair of neighbouring units.

Constructor Summary

Constructors

Constructor and Description
UnifiedDistanceMatrix(boolean individualDistances, DistanceMeasure distance) Simple constructor.

Method Summary

Methods

Modifier and Type	Method and Description
private double[][]	averageDistances(NeuronSquareMesh2D map) Computes the distances between a unit of the map and its neighbours.
double[][]	computeImage(NeuronSquareMesh2D map) Creates an image of the map.
private double[][]	individualDistances(NeuronSquareMesh2D map) Computes the distances between a unit of the map and its neighbours.

Methods inherited from class java.lang.Object

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

Field Detail

individualDistances

private final boolean individualDistances

Whether to show distance between each pair of neighbouring units.

distance

private final DistanceMeasure distance

Distance.

Constructor Detail

UnifiedDistanceMatrix

public UnifiedDistanceMatrix(boolean individualDistances,
                     DistanceMeasure distance)

Simple constructor.

Parameters:

individualDistances - If true, the 8 individual inter-units distances will be computed. They will be stored in additional pixels around each of the original units of the 2D-map. The additional pixels that lie along a "diagonal" are shared by two pairs of units: their value will be set to the average distance between the units belonging to each of the pairs. The value zero will be stored in the pixel corresponding to the location of a unit of the 2D-map.
If false, only the average distance between a unit and all its neighbours will be computed (and stored in the pixel corresponding to that unit of the 2D-map). In that case, the number of neighbours taken into account depends on the network's neighbourhood type.

distance - Distance.

Method Detail

computeImage

public double[][] computeImage(NeuronSquareMesh2D map)

Creates an image of the map.

Specified by:

computeImage in interface MapVisualization

Parameters:

map - Map.

Returns:

a 2D-array (in row major order) representing the property.

individualDistances

private double[][] individualDistances(NeuronSquareMesh2D map)

Computes the distances between a unit of the map and its neighbours. The image will contain more pixels than the number of neurons in the given map because each neuron has 8 neighbours. The value zero will be stored in the pixels corresponding to the location of a map unit.

Parameters:

map - Map.

Returns:

an image representing the individual distances.

averageDistances

private double[][] averageDistances(NeuronSquareMesh2D map)

Computes the distances between a unit of the map and its neighbours.

Parameters:

map - Map.

Returns:

an image representing the average distances.