fisheriris data and perceptron - matlab

Question

i want to apply the perceptron algorithm for fisheriris data and i was tried this code
function [ ] = Per( )
%PERCEPTON_NN Summary of this function goes here
% Detailed explanation goes here
%%%%%%%%%%%STEP ONE INPUT DATA PREPERATION
%N=3000;
load fisheriris
tr=50; %traning
te=50; %test
epochs =150;
data=meas;
%N = size(meas,1);
%species=nonomil(species)
%figure,plot(data_shuffeled(1,:),data_shuffeled(2,:),'rx');
%%%%%%%%%%%%%%%%%%%STEP TWO INTIALIZE WEIGHT
baise=1;
w=[baise; 1 ; 1;1 ; 1];
eta=0.9; %%learning rate
%%%%%%%%%%%%%%%%%%%%%%%% Rosen Blatt learning Algo
for epoch=1 : epochs
for i=1 : tr
x=[1;data(i,1);data(i,2);data(i,3);data(i,4)]; % input vector
N = size(species,i); %desiard output
y=w'*x; % y=actual output ,w'= transpose w , mmoken y=w.*x
%%%%%%%%%%%%%%% Activation function (hardlimit)(step fn)
y=1*(y>=0)+(-1)*(y<0); % da badl el if
%%%%%%%%%%% Error calcualtion %%%%%%%
err(i)=N-y;
%%%%%%%%%%%%%% update weight %%%%%%%%%5
wnew=w+ eta*err(i)*x;
w=wnew;
end
mse(epoch)=mean(err.^2);
end
%%%%%%%%%%%%%%%%%%%%%% step four classification (testing) %%%%%%%%%%%%%%%%%%5
hold on
for i=1 : te
%x=[1;data(i,1);data(i,2),data(i,3);data(i,4)];
x=[1;data(i,1);data(i,2);data(i,3);data(i,4)];
% d=data_shuffeled(3,i+tr);
N = size(species,i);
y=w'*x;
y=1*(y>=0)+(-1)*(y<0);
if (y==1)
plot(x(2),x(3),x(4),x(5),'rx');
elseif y==-1
plot(x(2),x(3),x(4),x(5),'r&');
end
end
hold off
if abs(N-y)>1E-6
testerro=testerro+1;
end
i wrote this code to make the perceptron algorithm with fisheriris data "meas" as input and species as "output"
any help in the code or any modify on this code .
Thanks .

Answer 1

First, did you know that MATLAB has something for neural network training called the Neural network toolbox?
Second, think data_shuffeled is your own function. There is something called randperm in MATLAB that you should use to shuffle your data.
Third, you want to avoid using for-loops when you can use vectors/matrices in MATLAB.
Instead of doing (for testing)
for i = 1:te,
....
end
You might want to do
X = [ones(te,1), data]; %X is [50x5] so each row of X is x'
y = X*w; %y is [50x1], X is [50x5], w is [5x1]
idx_p1 = y==1; %all the indices of y where y is +1
idx_m1 = y==-1; %all the indicies of y where y is -1
plot(X(idx_p1,1),y(idx_p1),'rx');
plot(X(idx_m1,1),y(idx_m1),'r&');
I don't know how you were using plot with 4-dimensional X so the above just plots with the first feature (column) of X.
Additionally, the training looks strange to me. For one, I don't think you should use N for both the size of data matrix meas and for the desired output. 'yhat' or 'ybar' is a better name. Also, if N is the desired output, then why is it size(species,i) where i loops through 1:50? species is a [150x1] vector. size(species,1) = 150. And size(species,x) where x is 2 to 50 will be 1. Are you sure you want this? Shouldn't it be something like:
yhat = -ones(50,1); %Everything is -1
yhat(strmatch('virginica,species)) = 1; %except virginicas which are +1