What is the following code doing?   global DEBUG;DEBUG=true;A=[ 1    -8     2; …     5     2     6; …     6    -4     4];b=[ 13; …    -6; …    14];P = [ 0, 0, 1;      1, 0, 0;      0, 1, 0 ];A = P*A;b = P*b;n = 3;debug_mat(“A: “, A);debug_mat(“b: “, b);C = A;for k=1:n    C(k,k) = 0;    x(k,1) = 0;    epsilon_a(k,1) = 1;endfor k=1:n    C(k,1:n) = C(k,1:n)/A(k,k);        d(k) = b(k)/A(k,k);enddebug_mat(“x: “, x);debug_mat(“C: “, C);debug_mat(“d: “, d);show = [‘No! ‘; ‘Yes!’];% from book, Fig 12.2, es=0.00001% is 0.0000001% called the ‘stop criterion’book_es = 0.0000001;for iter=1:50    % an nx1 column vector; should b zeros!    e = A*x – b;      % Griffis Method:  a little looser bound; cheat a little here    % this is also nx1 column vector of “err” bounds    err = ((n+1)*abs(A)*abs(x)+abs(b))*eps;    err_subm = (norm(A,1)*norm(x,1)+norm(b,1))*eps;        % important tests!  UNDERSTAND what these say.    x_is_floating_pt_valid = norm(e,1) ‘, repmat(‘%8.16f ‘, 1, n), ‘\n’];    fprintf(“%s\n”, entry);     fprintf(fmt1, A’);     fprintf(“\n”);end

Say is the square matrix,

This question concerns the coefficient of determination, of a linear regression of some raw data that is thought to depend on some raw data.  Consider that any fit we try to do gives an improvement over the straight average of the data.  How is bounded?

Say you have two equations

Iterative Methods—non-Linear ————————————–  

Gauss-Seidel can easily solve for a vector

Say we perform a linear least squares fit on the log of data such that and we know the constants .  What would be the “power fit” equivalent?

Iterative methods—Linear ———————————–  

What is the precise interpretation of a number in the structure matrix of the discriminant analysis output? 

What is l1-regularized logistic regression? What is the purpose of l1-regularization?