fminunc: Too many output arguments

Hi all, I can't find my bug in the following call of fminunc. The relevant section of my main file looks like this:

global W obs log_share price cost_shifter1 cost_shifter2 cost_shifter3 product_attrib1 product_attrib2 product_attrib3

opts    = optimset('display','iter-detailed','Diagnostics','on','TolFun',1e-10,'TolX',1e-10,'GradObj','on','DerivativeCheck','off','Hessian','user-supplied');
start   = beta_2SLS;

iter=1;
maxiter=2;
W=eye(6);

while iter<=maxiter
[betahat, fval] = fminunc(@(beta) moments(beta),start,opts);

W = varcov(betahat);
iter=iter+1;
end

That is, I call fminunc to minimize the function moments specified here:

function distance=moments(beta)

global W obs log_share price cost_shifter1 cost_shifter2 cost_shifter3 product_attrib1 product_attrib2 product_attrib3 

residual1=zeros(obs,1);
for i_obs=1:obs
           residual1(i_obs) = log_share(i_obs) - beta(1)-beta(2)*price(i_obs)- ...
           [beta(3), beta(4), beta(5)]*[product_attrib1(i_obs); product_attrib2(i_obs); product_attrib3(i_obs)];
end

moment=(1/obs)*residual1'*[cost_shifter1, cost_shifter2, cost_shifter3, product_attrib1, product_attrib2, product_attrib3]; 
% ( 1 X 6) vector

distance = moment*W*moment'; 
end

When I run the section main file, I get the error message: Error using moments Too many output arguments.

Error in @(beta)moments(beta)

Error in fminunc (line 285) [f,GRAD,HESS] = feval(funfcn{3},x,varargin{:});

Caused by: Failure in initial objective function evaluation. FMINUNC cannot continue

Does anybody see my error or give me a hint, how to debug this? Thanks a lot.

Answer 1

[Alan Weiss]

In your options you set GradObj to 'on' and Hessian to 'user-supplied'. Yet your objective function does not return the gradient and Hessian of your objective function. For details on how to supply these, see Including Gradients and Hessian. If you have a Symbolic Math Toolbox™ license, you can calculate these quantities automatically, rather than doing it by hand; see Using Symbolic Math with Optimization Toolbox Solvers.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[Florestan]

Thanks a lot!