Daniel@0: function net = som(nin, map_size)
Daniel@0: %SOM	Creates a Self-Organising Map.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %	NET = SOM(NIN, MAP_SIZE) creates a SOM NET with input dimension (i.e.
Daniel@0: %	data dimension) NIN and map dimensions MAP_SIZE.  Only two-
Daniel@0: %	dimensional maps are currently implemented.
Daniel@0: %
Daniel@0: %	The fields in NET are
Daniel@0: %	  type = 'som'
Daniel@0: %	  nin = number of inputs
Daniel@0: %	  map_dim = dimension of map (constrained to be 2)
Daniel@0: %	  map_size = grid size: number of nodes in each dimension
Daniel@0: %	  num_nodes = number of nodes: the product of values in map_size
Daniel@0: %	  map = map_dim+1 dimensional array containing nodes
Daniel@0: %	  inode_dist = map of inter-node distances using Manhatten metric
Daniel@0: %
Daniel@0: %	The map contains the node vectors arranged column-wise in the first
Daniel@0: %	dimension of the array.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	KMEANS, SOMFWD, SOMTRAIN
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: net.type = 'som';
Daniel@0: net.nin = nin;
Daniel@0: 
Daniel@0: % Create Map of nodes
Daniel@0: if round(map_size) ~= map_size | (map_size < 1)
Daniel@0:     error('SOM specification must contain positive integers');
Daniel@0: end
Daniel@0: 
Daniel@0: net.map_dim = length(map_size);
Daniel@0: if net.map_dim ~= 2
Daniel@0:     error('SOM is a 2 dimensional map');
Daniel@0: end
Daniel@0: net.num_nodes = prod(map_size);
Daniel@0: % Centres are stored by column as first index of multi-dimensional array.
Daniel@0: % This makes extracting them later more easy.
Daniel@0: % Initialise with rand to create square grid
Daniel@0: net.map = rand([nin, map_size]);
Daniel@0: net.map_size = map_size;
Daniel@0: 
Daniel@0: % Crude function to compute inter-node distances
Daniel@0: net.inode_dist = zeros([map_size, net.num_nodes]);
Daniel@0: for m = 1:net.num_nodes
Daniel@0:     node_loc = [1+fix((m-1)/map_size(2)), 1+rem((m-1),map_size(2))];
Daniel@0:     for k = 1:map_size(1)
Daniel@0: 	for l = 1:map_size(2)
Daniel@0: 	    net.inode_dist(k, l, m) = round(max(abs([k l] - node_loc)));
Daniel@0: 	end
Daniel@0:     end
Daniel@0: end