gradDesc.m

function [w,f] = gradDesc(funObj,w,maxEvals,varargin)
% Find local minimizer of differentiable function

% Step size
alpha = 1e-2;

% Evaluate initial function and gradient
[f,g] = funObj(w,varargin{:});
funEvals = 1;

while funEvals < maxEvals
    
    wold = w;
    if(funEvals == 1)
        %normal gradient descent to start
        w = w - alpha*g;
    else
        %BB step
        alpha = (s'*y)/(y'*y);
        w = w - alpha*g;
    end
    
    gold = g;
    [f,g] = funObj(w,varargin{:});
    funEvals = funEvals + 1;
    
    %BB update s and y
    s = w - wold;
    y = g - gold;
    
    % Print out how we are doing
    optCond = norm(g,'inf');
    fprintf('%6d %15.5e %15.5e %15.5e\n',funEvals,alpha,f,optCond);
    
    if optCond < 0.1
        fprintf('Solution found with optCond < 0.1\n');
        break;
    end
end