avq_dat.m

FISMAT accommodates different arithmetic operators, fuzzification and defuzzification algorithm, imp

echo off
% avq_dat.m
%
% Adaptive Vector Quantisation with an autoassociative two-layer feedforward
% neural network.
% For a pattern space consisting of 500 two-dimensional Gausian normal 
% distributed pattern vectors with variance 1 and centroids or modes at
% (-1,1), (1,1), (1,-1), (-1,-1).
% Different unsupervised learning algorithm can be applied to the problem.
%
% Reference:
% Kong, Kosko : 'Differential Competive Learning for Centroid Estimation
% and Phoneme Recognition'. 
% IEEE Transcactions on Neural Networks, Vol.2, No.1, Jan.1991
% pp.118-124
%
% FSTB - Fuzzy Systems Toolbox for MATLAB
% Copyright (c) 1993-1996 by Olaf Wolkenhauer
% Control Systems Centre at UMIST
% Manchester M60 1QD, UK
%
% 20-May-1994

dec=input('Generating new training pattern ? y/n [n] : ','s');
if dec=='y',
  % Generating 4x500 Gaussian-normal distributed pattern vectors, with
  % centroids or modes at (-1,1), (1,1), (1,-1), (-1,-1).
  spc=500;
  x1=randn(1,spc)-ones(1,spc);
  y1=randn(1,spc)+ones(1,spc);
  x2=randn(1,spc)+ones(1,spc);
  y2=randn(1,spc)+ones(1,spc);
  x3=randn(1,spc)+ones(1,spc);
  y3=randn(1,spc)-ones(1,spc);
  x4=randn(1,spc)-ones(1,spc);
  y4=randn(1,spc)-ones(1,spc);
  samples=[[x1';x2';x3';x4'],[y1';y2';y3';y4']];
end;
dec=input('Pattern vector plot ? y/n [n] : ','s');
if dec=='y', 
  plot(samples(:,1),samples(:,2),'.');
  title('training pattern vectors');
  ylabel('centroids: (-1,1)   (1,1)   (1,-1)    (-1,-1)');
end;
clear x1 x2 x3 x4 y1 y2 y3 y4
p=4;
c=1;
wii=1;
wij=-1;
disp('');
disp(['p=',num2str(p),'  c=',num2str(c),'  wii=',num2str(wii),'  wij=',...
num2str(wij)]);
dec=input('Any changes ? y/n [n] : ','s');
if dec=='y', 
  disp('type return to continue.'); 
  keyboard; 
end;

dec=input('new figure ? y/n [n] : ','s');
if dec=='y', 
  figure; 
else, clf; end;
%axis([-3 3 -3 3]);grid on; xlabel('x');ylabel('y');
%title('convergence of quantization vectors mj(t)');
%hold on

dec=input('Random mix of sample pattern ? y/n [y] : ','s');
if isempty(dec), dec=='y'; end;
if dec=='y',
  rands=randperm(length(samples));
end;
dec=input('UCL, DCL - algorithm ? u/d : ','s');
subplot(2,2,4); axis([-3 3 -3 3]); hold on
xlabel('x');ylabel('y'); title('quantisation vector convergence :');
%set(gca,'DrawMode','fast'); 
%set(gca,'EraseMode','none');
if dec=='u',
  [y2,m2x,m2y]=ucl(samples,p,c,wii,wij,rands); met='ucl';
elseif dec=='d',
  [y2,m2x,m2y]=dcl(samples,p,c,wii,wij,rands); met='dcl';
end;
t=1:1:length(samples);
subplot(2,2,2);plot(t,m2x,t,m2y);ylabel('M(2,2)=y   M(1,2)=x');grid on
subplot(2,2,1);plot(y2);ylabel('yt1(2)'); xlabel('training time');
%subplot(2,2,4);plot(m2y);ylabel('M(2,2) = y'); 
grid on
subplot(2,2,3);axis([0 1 0 1]);set(gca,'Visible','off');
text(0,0,['wii = ',num2str(wii)]);
text(0,0.2,['wij = ',num2str(wij)]);
text(0,0.4,['pattern = ',num2str(length(samples))]);
text(0,0.6,'ct = 0.1(1-t/pattern)');
text(0,0.8,['c = ',num2str(c)]);
text(0,1,[met,' algorithm']);

clear p c spc dec wii wij met 
% pack
echo on