In the course of implementing a Self-organising feature map (SOM) I needed
to initialise the weight vecorts of each of nodes in the map with random
numbers. I wrote the method randomWeights below for the class Som 


ArrayedCollection variableSubclass: #Som
	instanceVariableNames: 'width height'
	classVariableNames: ''
	poolDictionaries: ''
	category: 'NeuralNets'

randomWeights
	
	"initialise the weights of the neurones to random vectors"
	"JS 5/24/2005 23:14 Caution: all rendome vectors are the same. Why?"
	
	|randomGenerator |
	randomGenerator _ Random new.
	self do: [ :each |  1 to: each size do:
		[:index | each at: index put: randomGenerator next]].

It turns out that the weight vectors of each of the nodes in the SOM are set
to the SAME random vector, when I expected them to be all DIFFERENT. What is
wrong with this nested iteration?

The following trial in a workspace 


map _ Som new: 30
map width:6 height:5 inputs: 4
count _1.
1 to: 30 do: [ :each | (map at: each) at: 1 put: count. Count _ count + 1]. 

sets the first weight of each of the 30 nodes to 30 !!!  I would expect them
to be set to 1, 2, ... 30.

Please help.

Thanks

Joaquin

  



=====================================================
Dr. Joaquin Sitte, Associate Professor.
Head, Smart Devices Lab
School of Software Engineering and Data Communication
Faculty of Information Technology
Queensland University of Technology
GPO Box 2434, Brisbane, Q 4001
Australia
Phone +61 7 3864 9325
Fax +61 7 3864 9390
e-mail j.sitte at qut.edu.au 
homepage http://www.fit.qut.edu.au/~sitte