how to set the dimensions for a SOM map?

While using SOM for clustering an image using the function net=selforgmap([d1,d2]) How to set the values for d1 and d2?

Answer 1

[Greg Heath]

1. Plot the projection of the data on the dominant PCA plane to get a "good feel" for the data:

a. lookfor 'principal component'

b. Test on the iris_dataset example

2. Create and train a SOM net using defaults:

   net = selforgmap;

3. Look at the 6 default training plots. PLOTSOMND, PLOTSOMHITS, and PLOTSOMPOS are probably the most useful for estimating a better choice for [d1,d2].

Hope this helps.

Thank you for formally accepting my answer.

[Greg Heath]

close all, clear all, clc, plt=0;

% x = iris_dataset;

x = simplecluster_dataset;

whos

[ I N ] = size(x) % [ 2 1000 ]

' 1. PROJECT THE DATA ON THE DOMINANT PRINCIPAL COMPONENT PLANE '

plt = plt+1; figure(plt)

plot(x(1,:),x(2,:),'o')

'2. CREATE AND TRAIN A MAP USING DEFAULTS'

net = selforgmap; %default is 8X8

net.IW{:} % Empty matrix: 0-by-64

net.LW{:} % []

net.b{:} % []

net = train(net,x);

net.IW % [64x2 double]

net.LW % { [] }

net.b % { [] }

'3. VIEW THE 6 DEFAULT SOM PLOTS. THE LAST 3 ARE THE MOST USEFUL FOR EVALUATING PERFORMANCE: '

plt = plt+1; figure(plt)

plotsomtop(net) % SOM Topology

plt = plt+1; figure(plt)

plotsomnc(net) % SOM Neighbor Connections

plt = plt+1; figure(plt)

plotsomplanes(net) % SOM Input Planes

plt = plt+1; figure(plt)

plotsomnd(net,x) %(The Neighbor Weight Distances)

plt = plt+1; figure(plt)

plotsomhits(net,x) %(SOM Sample Hits)

plt = plt+1; figure(plt)

plotsompos(net,x) %(SOM Weight Positions))

'4. ASSIGN EACH INPUT TO ONE OF THE d1*d2 SOM CENTERS'

y = net(x); % 64 X 1000

classes = vec2ind(y); % 1 X 1000

example = classes(1:20)