Predicting Mortality of ICU Patients: The PhysioNet/Computing in Cardiology Challenge 2012 1.0.0

function alpha = mb_nocLineSearch(f,gradF,x,dir,slope0,of)

% implementation of strong wolfe line search
% 
% original c++ code by m.siggel@dkfz.de
% import to matlab by m.bangert@dkfz.de
% reference: nocedal: numerical optimization 3.4 line search methods
%
% f       - function handle of objective function
% gradF   - function handle of gradient
% x       - current iterate
% dir     - search direction
% slope0  - gradient of function at x in direction dir
% alphaLo - alpha at lower objective function value for zoom
% alphaHi - alpha at higher objectvie function value for zoom
% of_0
% oflo

% parameter for sufficient decrease condition
c1 = 0.001;
% parameter for curvature condition
c2 = 0.1;

if c1 > c2
    error('c1 > c2\n');
end

alphaMax = 100;
alpha    = 1;

alpha_0  = 0;
alpha_1  = alpha;

of_x = of;
of_0 = of;
iter = 0;

while 1

    xc     = x+alpha_1*dir;
    of     = f(xc);
    slopec = gradF(xc)'*dir;
    
    % check if current iterate violates sufficient decrease
    if  (of > of_0 + slope0*c1*alpha_1) || ((of >= of_x ) && (iter > 0))
        % there has to be an acceptable point between alpha_0 and alpha_1
        % (because c1 > c2)
        alpha = nocZoom(f,gradF,x,dir,slope0, alpha_0, alpha_1,of_0,of_x ,c1,c2);
        break;
    end
    % current iterate has sufficient decrease, but are we too close?
    if(abs(slopec) <= -c2*slope0)
        % strong wolfe fullfilled, quit
        alpha = alpha_1;
        break;
    end
    % are we behind the minimum?
    if (slopec >= 0)
        % there has to be an acceptable point between alpha_0 and alpha_1
        alpha = nocZoom(f,gradF,x,dir,slope0,alpha_1 , alpha_0,of_0,  of,c1,c2);
        break;
    end

    alpha_0 = alpha_1;
    alpha_1 = min(alphaMax, alpha_1*3);
    of_x = of;

    iter = iter + 1;
end

end

function alpha = nocZoom(f,gradF,x,dir,slope0,alphaLo,alphaHi,of_0,ofLo,c1,c2)
% this function is only called by mb_nocLineSearch - everything else does
% not make sense!

while 1
    
    alpha = (alphaLo+alphaHi)/2;
    xc    = x + alpha*dir;
    of    = f(xc);
    
    if of > of_0 + c1*alpha*slope0 || of >= ofLo
        % if we do not observe sufficient decrease in point alpha, we set
        % the maximum of the feasible interval to alpha
        alphaHi = alpha;       
    else
        slopec = gradF(xc)'*dir;
        % strong wolfe fullfilled?
        if abs(slopec) <= -c2*slope0
            return;
        end
        if slopec*(alphaHi-alphaLo) >= 0 % if slope positive and alphaHi > alphaLo  
            alphaHi = alphaLo;
            alphaLo = alpha;
            ofLo    = of;
        end
    end
    
end

end