Package smile.vq

Class GrowingNeuralGas

java.lang.Object

smile.vq.GrowingNeuralGas

All Implemented Interfaces:

Serializable, VectorQuantizer

public class GrowingNeuralGas
extends Object
implements VectorQuantizer

Growing Neural Gas. As an extension of Neural Gas, Growing Neural Gas can add and delete nodes during algorithm execution. The growth mechanism is based on growing cell structures and competitive Hebbian learning.

Compared to Neural Gas, GNG has the following distinctions:

• The system has the ability to add and delete nodes.
• Local Error measurements are noted at each step helping it to locally insert/delete nodes.
• Edges are connected between nodes, so a sufficiently old edges is deleted. Such edges are intended place holders for localized data distribution.
• Such edges also help to locate distinct clusters (those clusters are not connected by edges).

References

1. B. Fritzke. A growing neural gas network learns topologies. NIPS, 1995.

See Also:

• KMeans
• NeuralGas
• NeuralMap
• Serialized Form

Field Summary

Fields inherited from interface smile.vq.VectorQuantizer

OUTLIER

Constructor Summary

Constructors

Constructor	Description
GrowingNeuralGas(int d)	Constructor.
GrowingNeuralGas(int d, double epsBest, double epsNeighbor, int edgeLifetime, int lambda, double alpha, double beta)	Constructor.

Method Summary

Modifier and Type	Method	Description
Neuron[]	neurons()	Returns the neurons in the network.
double[]	quantize(double[] x)	Quantize a new observation.
void	update(double[] x)	Update the codebook with a new observation.

Methods inherited from class java.lang.Object

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

Constructor Details

GrowingNeuralGas

public GrowingNeuralGas(int d)

Constructor.

Parameters:

d - the dimensionality of signals.

GrowingNeuralGas

public GrowingNeuralGas(int d,
 double epsBest,
 double epsNeighbor,
 int edgeLifetime,
 int lambda,
 double alpha,
 double beta)

Constructor.

Parameters:

d - the dimensionality of signals.

epsBest - the learning rate to update best matching neuron.

epsNeighbor - the learning rate to update neighbors of best matching neuron.

edgeLifetime - the maximum age of edges.

lambda - if the number of input signals so far is an integer multiple of lambda, insert a new neuron.

alpha - decrease error variables by multiplying them with alpha during inserting a new neuron.

beta - decrease all error variables by multiply them with beta.

Method Details

update

public void update(double[] x)

Description copied from interface: VectorQuantizer

Update the codebook with a new observation.

Specified by:

update in interface VectorQuantizer

Parameters:

x - a new observation.

neurons

public Neuron[] neurons()

Returns the neurons in the network.

Returns:

the neurons in the network.

quantize

public double[] quantize(double[] x)

Description copied from interface: VectorQuantizer

Quantize a new observation. Returns Optional.empty if the observation is noise.

Specified by:

quantize in interface VectorQuantizer

Parameters:

x - a new observation.

Returns:

the quantized vector.