selforgmap

Self-organizing map

Syntax

selfOrgMap = selforgmap(dimensions)

selfOrgMap = selforgmap(dimensions,coverSteps,initNeighbor,topologyFcn,distanceFcn)

Description

Self-organizing maps learn to cluster data based on similarity, topology, with a preference (but no guarantee) of assigning the same number of instances to each class.

You can use self-organizing maps to cluster data and to reduce the dimensionality of data. They are inspired by the sensory and motor mappings in the mammal brain, which also appear to automatically organizing information topologically.

selfOrgMap = selforgmap(dimensions) takes a row vector of dimension sizes and returns a self-organizing map.

selfOrgMap = selforgmap(dimensions,coverSteps,initNeighbor,topologyFcn,distanceFcn) takes a row vector of dimension sizes and also a number of training steps for initial covering, an initial neighborhood size, a layer topology function, and a neuron distance function, and returns a self-organizing map.

Examples

Use Self-Organizing Map to Cluster Data

This example shows how to use a self-organizing map to cluster a simple set of data.

x = simplecluster_dataset;
net = selforgmap([8 8]);
net = train(net,x);

view(net)

y = net(x);
classes = vec2ind(y);

Input Arguments

dimensions — Dimension sizes

[8 8] (default) | row vector

Dimension sizes, specified as a row vector.

coverSteps — Initial covering steps

100 (default) | scalar

Number of training steps for initial covering of the input space, specified as a scalar.

initNeighbor — Initial neighborhood size

3 (default) | scalar

Initial neighborhood size, specified as a scalar.

topologyFcn — Topology function

'hextop' (default) | 'randtop' | 'gridtop' | 'tritop'

Layer topology function, specified as a topology function.

distanceFcn — Distance function

'linkdist' (default) | 'dist' | 'mandist'

Neuron distance function, specified as a distance function.

Output Arguments

selfOrgMap — Self-organizing map

network object

Self-organizing map, returned as a network object.

Version History

Introduced in R2010b