SVM MATLAB implementation Error plotting decision boundary

Question

I am trying to do binary classification in MATLAB but the following code throws an error at the end.

   load('ex6data1.mat');

% Plot training data
plotData(X, y);



fprintf('Program paused. Press enter to continue.\n');
pause;

model=fitcsvm(X,y);
visualizeBoundaryLinear(X,y,model); //error shows up here i guess


fprintf('Paused');




Error window
   Error using subsref
No appropriate method, property, or field 'w' for class
'ClassificationSVM'.

Error in classreg.learning.internal.DisallowVectorOps/subsref (line
21)
                [varargout{1:nargout}] = builtin('subsref',this,s);

Error in visualizeBoundaryLinear (line 7)
w = model.w;

Error in Untitled2 (line 18)
visualizeBoundaryLinear(X,y,model);

Note:Y is 1 for positive example and -1 for negative.

What error does matlab throw? Can you copy it under your code please? — BillBokeey
What does visualizeBoundaryLinear look like? The error happens in that function, but visualizeBoundaryLinear is not part of the native MATLAB distribution. — rayryeng
What i want to do is plot the linear decision boundary obtained by the SVM..why is the visualizeboundarylinear function throwing an error..isnt it supposed to help me do what i want? — Sridhar Thiagarajan
Yeah it's supposed to do that, but the code is happening in that function... how can we help you if we don't see what the code is doing? — rayryeng

GJStein GJStein · Accepted Answer · 2015-10-16T14:14:57

If you look at the MATLAB documentation for fitcsvm you'll find that there is no property w, which is what's giving you your error.

You need to calculate the weights w yourself, since MATLAB is solving the dual form of the SVM. More details can be found here. Take a look at this reference if you want to learn more. You can use the following formula:

w = zeros( size(x(1,:)) );
for i=1:N
    w = w + alpha(i)*y(i)*x(i,:);
end

You can calculate the w vector using a combination of the alpha's, which are returned in your model, and your data.

SVM MATLAB implementation Error plotting decision boundary

1 Answers