############################################################################## function W = overoptimize(s, sigma, dim1, dim2, alphamax, c1, c2, converge)
% This is the main function used to optimize the objective function % % Parameters % ---------- % s: struct('A',[],'c',[]), % sigma: kernel width in Gaussian kernel, % dim1: original dimension of data, % dim2: reduced dimension of data, % alphamax: the maximum alpha, % c1: constant for first wolfe condition, % c2: constant for second wolfe condition, % converge: the threshold for convergence.
% Returns % ------- % W: projection matrix with dimensions dim1 x dim2
For the struct s, s(index).c is the parameter in objective function for xi and xj. s(index).A is the d-by-d rank one matrix defined as (xi-xj)T(xi-xj).
##############################################################################
function [w valuevector]= optimizeW(s, sigma, w, W, alphamax, c1, c2,... converge,valuevector)
% This function optimize the objective function w.r.t vector w
% Parameters % ---------- % s: struct('A',[],'c',[]), % sigma: kernel width in Gaussian kernel, % w: w is the new vector that need to be optimized % W: W is the matrix consisting of existing vectors,note that w should % be orthogonal to the column vectors in W % alphamax: maximum value of alpha in line search % c1: constant for first wolfe condition, % c2: constant for second wolfe condition, % converge: threshold for convergence.
% Returns % ------- % w: updated value of the vector optimizer % valuevector: store the values of objective function in each step