neingeist
/
coursera-ml-007-exercises
Archived
1
0
Fork 0
Code Releases Activity

Regularized NN gradient

Browse Source
master
neingeist 10 years ago
parent bdecab8cf8
commit eccdcc0d81
1 changed files with 7 additions and 3 deletions
Show Stats Download Patch File Download Diff File

10
ex4/nnCostFunction.m
Unescape Escape View File

@ -124,12 +124,16 @@ Theta1_grad = D_1;
% Note: Theta1/2 are matrixes here, we want all their rows, but skip their % Note: Theta1/2 are matrixes here, we want all their rows, but skip their
% first column (not regularizing the bias term). % first column (not regularizing the bias term).
regularization_term = lambda/(2*m) * ... J_regularization_term = lambda/(2*m) * ...
(sum(sum(Theta1(:,2:end).^2)) ... (sum(sum(Theta1(:,2:end).^2)) ...
+ sum(sum(Theta2(:,2:end).^2))); + sum(sum(Theta2(:,2:end).^2)));
assert(size(regularization_term) == [1 1]); assert(size(J_regularization_term) == [1 1]);
J += J_regularization_term;
J += regularization_term; Theta2_grad_regularization_term = lambda/m * [zeros(size(Theta2, 1), 1) Theta2(:,2:end)];
Theta1_grad_regularization_term = lambda/m * [zeros(size(Theta1, 1), 1) Theta1(:,2:end)];
Theta2_grad += Theta2_grad_regularization_term;
Theta1_grad += Theta1_grad_regularization_term;
% ------------------------------------------------------------- % -------------------------------------------------------------