How Quadratic discriminant is separtaing data in projected data space?
1 view (last 30 days)
Show older comments
I have a 4 dimensional data space. I did PCA on the data and tried to separate the data by using FLDA. I understood FLDA projects the data in one dimension and the parameters are adjusted to give a specific choice of projection direction with maximum separation.What I understood from example in matlab FLDA, that the visualization of the separator/discriminator separating two classes do not let us visualize the separation in this FLDA space. I tried projecting the data into one D space and wanted to see the separation. But in case of Quadratic discriminator, there are two discriminator passing in between the clusters and this discriminator is not a conic. I am not sure what I understood is correct and how does these two discriminant classifies my two class problem.
My code does following:
X = Data(:,1:2);
Z = {};
for i = 1:208
Z{i}='flat';
end
for i = 209:416
Z{i} = 'nodules';
end
T = cat(2,Z{:});
T = Z';
cls = ClassificationDiscriminant.fit(X,T,'DiscrimType','quadratic');
hold all;
K = cls.Coeffs(1,2).Const;
L = cls.Coeffs(1,2).Linear;
Q = cls.Coeffs(1,2).Quadratic;
f = @(x1,x2) K + L(1)*x1 + L(2)*x2 + Q(1,1)*x1.^2 + ...
(Q(1,2)+Q(2,1))*x1.*x2 + Q(2,2)*x2.^2;
% Projecting data into FLDA space
y = f(X(:,1),X(:,2));
%Orthogonal projections of data on single dimension
plot(y(:,1:208),'*r')
hold on
plot(y(:,209:416),'*g')
%drawing the quadratic separator
h2 = ezplot(f,[-90 450 -60 20]);
set(h2,'Color','b','LineWidth',2)
0 Comments
Answers (0)
See Also
Categories
Find more on Dimensionality Reduction and Feature Extraction in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!