Trying to plot error

I'm trying to run this code and get plot of the error as a function of epsilon but im getting blank plot, someone can help?

    % 
clear all;close all;clc
epsilon=0.0001;
X = randn(2,1000); % make some input data
w = [1;1]; % define a linear function
n = randn(1,1000)*0.1; % produce some noise
Y = w'*X + n; % produce outputs
% scatter3(X(1,:),X(2,:),Y);
for iteration = 1 : 500 %in practice: until stopping
%criterion satisfied
grad = 2*sum(repmat(w'*X-Y,size(X,1),1).*X,2);
w = w - epsilon * grad;
err = sum((Y - w'*X).^2) %just to check
end

plot(err,epsilon)

end

Answer 1

[Star Strider]

You never change ‘epsilon’ in your code, so it remains a single scalar value. You need to create a vector from it if you want to plot with respect to it.

If you want to create a vector from ‘err’, subscript it:

err(iteration) = sum((Y - w'*X).^2); %just to check

[rivaldo rivaldo]

i'm still getting blank plot...there is any other way?

[Star Strider]

You will get a blank plot unless you create a vector for °epsilon’ that is in some way related to it.

After you create a vector for ‘err’, I would just plot it as:

plot(err)
grid

[rivaldo rivaldo]

now it works thanks, but this solution is show me plot of the error as a function of the iterations. im looking for plot for the error as a function of epsilon.

[Star Strider]

You never changed or updated ‘epsilon’, so it remains a single scalar value. Until you create it as a vector, you cannot plot with respect to it. It has to be a vector the same size as ‘err’ to work with the plot function.

You can plot it with respect to ‘w’ if you change your code in the loop to:

w = w - epsilon * grad;
wv(iteration) = w;
err(iteration) = sum((Y - w'*X).^2); %just to check

then after the loop:

plot(wv, err)

[rivaldo rivaldo]

i'm getting error

 Subscripted assignment dimension mismatch.

[Guillaume]

Make sure to preallocate w and err before the loop. w is a vector so wv(iteration) = w is indeed an error.

numiterations = 500;
wv = zeros(numiterations , 2);
er =zeros(numiterations, 1);
for iteration = 1:numiterations
   grad = 2*sum(repmat(w'*X-Y,size(X,1),1).*X,2);
   w = w - epsilon * grad;
   wv(iteration, :) = w;
   err(iteration) = sum((Y - w'*X).^2) %just to check
end

Still no idea what you want to plot against. It can't be epsilon which never changes or w which is a 1x2 vector that changes at each iteration.

[rivaldo rivaldo]

i want to plot the error(err) against epsilon

[Star Strider]

Please see my Answer and previous Comments.

You must create epsilon as a vector in order to plot with respect to it.

Try this:

epsilonv = ones(size(err))*epsilon;                         % Create Vector For ‘epsilon’
plot(epsilonv, err, 'p')

[njj1]

But epsilon is not changing.

What if you did this instead:

clear all;close all;clc
epsilon=0.0001:0.0001:0.0009; %nine entries into epsilon
X = randn(2,1000); % make some input data
w = [1;1]; % define a linear function
n = randn(1,1000)*0.1; % produce some noise
Y = w'*X + n; % produce outputs
numiterations = 500;
wv = zeros(numiterations , 2);
err =zeros(numiterations, 1);
for j = 1:numel(epsilon)
   e = epsilon(j);
   for iteration = 1:numiterations
      grad = 2*sum(repmat(w'*X-Y,size(X,1),1).*X,2);
      w = w - e * grad;
      wv(iteration, :) = w;
      err(iteration,j) = sum((Y - w'*X).^2); %just to check
   end
end
plot(epsilon,err(end,:)); %plot err at the final iteration against epsilon