To compute cost and gradient for a logistic regression problem:the function returns only zeroes even though the correct value is being computed(I know it because I executed the function body separately in command line and checked)

function [J,grad]=costFunction(theta, X, y)

m = length(y); 
z=X*theta;
h=sigmoid(z);
H=h.';
h1=log(H)*(-y);
h2=(log(1-H))*(1-y);
J=(h1-h2)/m;
s=h-y;
s=s.';
for i=1:3,
    
    temp=(s)*X(:,i);
    grad(i)=temp(1);
end

grad=grad./m;

% =============================================================

end

[Sahithi Kanumarlapudi]

The way the function is implemented looks correct. There might be something wrong where this function is being called.

Could you provide me the dataset with which you are trying to call this function so that I can verify.

[spurgin waters]

How does the ex2.m dataset get called into the function?

Answer 1

[Antoine Jonathan]

There is a problem somewhere in your data manipulation. the cost function is right.

I started from the beginning for the gradients, now it works:

Z2 = X * theta;
A2 = sigmoid(Z2);
D = (A2 - y);
grad(1) = D' * X(:,1);
grad(2) = D' * X(:,2);
grad(3) = D' * X(:,3);

grad = grad/m;