% compet7.m
%
clf reset
pausetime = 0.05;
% PROBLEM DEFINITION
%===================

peec7

P=normc(PE);

% INITIALIZE NETWORK ARCHITECTURE
%================================
% Set input layer size R and neuron layer size S.
[R,Q] = size(P); S = 5;

  %W0 = randnc(S,R);
W0 =[-0.5953   -0.7837;
     -0.2435    0.0032;
     -0.8005    0.5992;
      0.6311   -0.7735;
      0.9771   -0.1390];
%=======
plot(cos(0:.1:2*pi),sin(0:.1:2*pi),'--')
axis('equal')
hold on
plotv(P);
h=plot(W0(:,1)',W0(:,2)','.');
title ('Input Vectors (lines) & Weight Vectors (+)');
xlabel('P(1,q)   W(i,1)');
ylabel('P(2,q)   W(i,2)');

% TRAIN THE NETWORK
%==================
% TRAINING PARAMETERS
disp_freq = 10;
max_cycle = 500;
lp.lr = 0.05;

% NETWORK PARAMETER
W = W0;

LW = W;
for cycle=1:max_cycle
  
  % PRESENTATION PHASE
  q = fix(rand*Q) + 1;
  A = compet(W*P(:,q));

  % LEARNING PHASE
  i = find(A == 1);
  dW = learnk(W,P(:,i),[],[],A,[],[],[],[],[],lp,[]);
  
  W = W + dW;
  % DISPLAY PROGRESS

  if rem(cycle,disp_freq) == 0
    delete(h)
    h = plot(W(:,1)',W(:,2)','+');
    LW = W;
    pause2(pausetime);
    end
  
  end

pause
hold off

% SUMMARIZE RESULTS
%==================

fprintf('\nFINAL NETWORK VALUES:\n');
W
fprintf('Trained for %.0f cycles.\n',max_cycle)

YE=hardlim(W*P-.95)

for i=1:5,
W2=sqrt(W(i,1)^2+W(i,2)^2);
WE(i,:)=[W(i,1) W(i,2)]*W2;
end

P
PE
WE
YE