Completed:
Coursera Week 5 Lectures
Review Coursera weeks 1-4 material
Complete EX 4
TODO:
Lessons Learned:
%"unroll" matrices into vector
%[matrix1(:); matrix2(:); matrix3(:); ] --> combine three matrices into one long vector
thetaVector = [ Theta1(:); Theta2(:); Theta3(:); ]
Theta1 = reshape(thetaVector(1:110),10,11) --> columns 1:110 turned into 10 X11 Theta2 = reshape(thetaVector(111:220),10,11) Theta3 = reshape(thetaVector(221:231),1,11)
input to fminunc is the unrolled vector
use matrices for forward and backward propagation
Use unrolled vector for calculation of D
initial_theta: use randomize vector (not zeros(n, 1))
Theta1 = rand(10,11) * (2 * INIT_EPSILON) - INIT_EPSILON; Theta2 = rand(10,11) * (2 * INIT_EPSILON) - INIT_EPSILON; Theta3 = rand(1,11) * (2 * INIT_EPSILON) - INIT_EPSILON; X^2 = X*X' X^2 != X*X
Default network architecture: 1 hidden layer
EX 4
∑∑A∗B= trace (A'*B ) = trace(B'*A)
% X(:,2:end) --> select all except first column of the matrix
R1 = Theta1(:,2:end) * Theta1(:,2:end)';
EX 5
set theta(1)=0;
Example of High Bias