machine learning - Not getting correct value of cost function in neural network classifier with pre calculated weights. [Octave] -

i'm facing problem in programming assignment. formula calculating cost function neural network is:

in case, h 5000x10 matrix , y 5000x1 vector. i'm calculating using following code:

    x = [ones(m, 1), x];     a2 = sigmoid(x * theta1');     a2new = [ones(m, 1), a2];     h = sigmoid(a2new * theta2');     %x matrix of training examples, theta1 , theta2 precalculated weights, a2 hidden layer.      k = num_labels; % number of classes     = 1:k;         c = y==k;         j = (-1/m)*((c' * log(h(:,i)))' + ((1-c)' * log(1-h(:,i)))');     end 

the answer around 0.23 i'm getting 9. can tell me wrong in code?

for code, think should y == i rather y==k in for-loop.

for = 1:k;     c = y==i;       j = (-1/m)*((c' * log(h(:,i)))' + ((1-c)' * log(1-h(:,i)))'); end 

and

second, way vectorize y seems not correct, should (1:num_labels)' == y' using vectorization implementation.