%Script file:  Iris classification using Matlab's RBF routines.

numiters=20;  %Number of widths of Gaussians
numtrain=120; %Number of training points to use

load IrisData  %From earlier, loads X and Y, 150 points

eg=0.1;             %Error goal
sp=linspace(0.05,1.6,numiters);  %Gaussian Widths

[NumPoints,Dim]=size(X);
temp=randperm(NumPoints);
P=X(temp(1:numtrain),:)';
T=Y(temp(1:numtrain),:)';

warning off  %We expect some warnings about invertibility

for i=1:numiters
%print the iteration so you can keep track:
  i
  net=newrb(P,T,eg,sp(i));
  z=sim(net,X');
  e(i)=0;

  [c(i),b]=size(net.IW{1});  %c is the number of weights used

  for k=1:150
    [x,j]=max(z(:,k));
    if Y(k,j)~=1
      e(i)=e(i)+1;           %e is the error in classification
    end
  end
  W(i)=max(max(abs(net.LW{2,1})));  %Max weight size
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%  Plotting routines
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

figure(1)
plot3(X(1:50,1),X(1:50,2),X(1:50,3),'r*')
hold on
plot3(X(51:100,1),X(51:100,2),X(51:100,3),'b^')
plot3(X(101:150,1),X(101:150,2),X(101:150,3),'go')
hold off
figure(2)
plot(sp,e)
title('Error versus Gaussian Width')
figure(3)
plot(sp,c)
title('Number of centers from OLS versus Gaussian Width');
figure(4)
semilogy(sp,W)
title('Maximum Weight versus Gaussian Width');