| [W_out, p_out]=som_1d(P, nsofm, epochs, phas, W) |
function [W_out, p_out]=som_1d(P, nsofm, epochs, phas, W)
% function W=som_1d(P, nsofm, lri, epochs, W)
%
% P - patterns
% nsofm - number of neurons
% epochs - number of epochs to train for
% phas - either
% two element vector for initialization phase
% [initial_learning_rate initial_neighbourhood_size]
% these values will decrease exponentially during training
% four element vector for convergence phase
% [initial_learning_rate final_learning_rate ...
% initial_neighbourhood_size final_neighbourhood_size]
% these values will decrease linearly during training
% W - initial weight values (optional)
%
% W_out - final weights
% p_out - parameters at the end of training (learning rate, neighbourhood size)
%
% Hugh Pasika 1997
if nargin == 4, W=randn(nsofm,2)*.1; end
mx=min(P(:,1));mxx=max(P(:,1));
my=min(P(:,2));myx=max(P(:,2));
n = 1:epochs;
if length(phas) == 2,
so = phas(2);
t1 = epochs./(log(so+1));
sigmas = so*exp(-n/t1);
lrs = phas(1)*exp(-n/epochs);
else
sigmas = (phas(3)-phas(4))*fliplr(n)/epochs+phas(4);
lrs = (phas(1)-phas(2))*fliplr(n)/epochs+phas(2);
end
[tpr tpc]=size(P);
ax=(max(abs([max(P) min(P)]))); % outer boundary for graphing
for epoch=1:epochs
Ps=shuffle(P);
sigma=sigmas(epoch);
lr = lrs(epoch);
disp([epoch lr sigma])
for i=1:tpr, % train for one epoch
%%%%%%%%%%% determine winner
t=Ps(i,:); Dup = t(ones(nsofm,1),:);
[toss, winner] = min(sum((Dup-W)'.^2));
mask = [-(winner-1):1:nsofm-winner];
m = (1/(2*pi*sigma^2)*exp((-mask.^2/(2*sigma^2))));
m = m/max(m);
m = [m' m'];
W = W + lr*m.*(Dup - W);
end
figure(1)
clf
plot(W(:,1),W(:,2))
axis([mx mxx my myx])
drawnow
W_out = W;
p_out = [lr(length(lr)) sigmas(length(sigmas))];
end
|
|
