plotting the Eigenvectors correctly in Matlab - matlab

Question

I'm trying to plot the caculated Eigenvectors of 2D dataset, here the script I've wrote for that:
clear ;
s = [2 2]
set = randn(200,1);
x = normrnd(s(1).*set,1)+3
x = zscore(x) % Standardize
y = normrnd(s(1).*set,1)+2
y= zscore(y)%Standardize
x_0 = mean(x)
y_0 = mean (y)
c = linspace(1,100,length(x)); % color
scatter(x,y,100,c,'filled')
xlabel('1st Feature : x')
ylabel('2nd Feature : y')
title('2D_dataset')
grid on
% gettign the covariance matrix
covariance = cov([x,y])
% getting the eigenvalues and the eigenwert
[eigen_vector, eigen_values] = eig(covariance)
eigen_value_1 = eigen_values(1,1)
eigen_vector_1 =eigen_vector(:,1)
eigen_value_2 = eigen_values(2,2)
eigen_vector_2 =eigen_vector(:,2)
% ploting the eigenvectors !
hold on
x_0 = repmat(x_0,size(eigen_vector_2,1),1);
y_0 = repmat(y_0,size(eigen_vector_1,1),1);
quiver(x_0, y_0,eigen_vector_2*(eigen_value_2),eigen_vector_1*(eigen_value_1),'-r','LineWidth',5)
and here is the result I'm getting:
I've double checked the math, the values are correct, but the plot is a mess !
Any idea what I'm missing in the plot of the 2 vectors ?
thanks in advance !

Answer 1

In your code, replace this part:
covariance = cov([x,y])
% getting the eigenvalues and the eigenwert
[eigen_vector, eigen_values] = eig(covariance)
eigen_value_1 = eigen_values(1,1)
eigen_vector_1 =eigen_vector(:,1)
eigen_value_2 = eigen_values(2,2)
eigen_vector_2 =eigen_vector(:,2)
% ploting the eigenvectors !
hold on
x_0 = repmat(x_0,size(eigen_vector_2,1),1);
y_0 = repmat(y_0,size(eigen_vector_1,1),1);
quiver(x_0, y_0,eigen_vector_2*(eigen_value_2),eigen_vector_1*(eigen_value_1),'-r','LineWidth',5)
with the following code:
covariance = cov([x,y]);
[eigen_vector, eigen_values] = eig(covariance);
eigen_vector_1 = eigen_vector(:,1);
eigen_vector_2 = eigen_vector(:,2);
d = sqrt(diag(eigen_values));
hold on;
quiver(x_0,y_0,eigen_vector(1,2),eigen_vector(2,2),d(2),'k','LineWidth',5);
quiver(x_0,y_0,eigen_vector(1,1),eigen_vector(2,1),d(1),'r','LineWidth',5);
hold off;
Does this produces what you are looking for? It looks much more coherent to me...

Answer 2

You are plotting the two components of one eigenvector as the x component of two vectors, and the other eigenvector as the y components.
[eigen_vector, eigen_values] = eig(covariance)
eigen_x = eigen_vector(1,:);
eigen_y = eigen_vector(2,:);
scale = diag(eigen_vector)'; % not sure what the output orientation is
% ploting the eigenvectors !
hold on
x_0 = repmat(x_0,size(eigen_vector_2,1),1);
y_0 = repmat(y_0,size(eigen_vector_1,1),1);
quiver(x_0, y_0,eigen_x.*scale,eigen_y.*scale,'-r')
Actually, because they are orthonormal, slicing the matrix the other way does not change much. But your scaling is changing the angles of the vectors, not just thier length, because of what I mention above.