nonlcon : Too many input arguments.

Hi everybody, I have a problem with non linear constraints in fmincon. I get this error "Too many input arguments".

The nonlcon function is very simple:

function [c,ceq] = covposdef(x)

c = -x(3)*x(5)+x(4)^2;

ceq = [];
end

The function to optimize is a likelihood of a bivariate normal density. Any suggestion?

[Matt J]

Are you sure it's coming from nonlcon? Copy/paste the complete error message.

[mklcst mklcst]

Error using covposdef
Too many input arguments.

Error in fmincon (line 718)
        [ctmp,ceqtmp] = feval(confcn{3},X,varargin{:});

Error in meancovarianceest (line 16)
theta = fmincon(@meancov,iv,[],[],[],[],lb,ub,@covposdef,options,data);

Caused by:
    Failure in initial user-supplied nonlinear constraint function evaluation. FMINCON
    cannot continue.

Answer 1

[Matt J]

I think the problem is that FMINCON thinks your fixed "data" is meant to be passed both to meancov and covposdef. If you intend only to pass it to meancov, call FMINCON using the following more modern syntax

theta = fmincon(@(x) meancov(x,data),iv,[],[],[],[],lb,ub,@covposdef,options);

[mklcst mklcst]

Thank you, it's working but I obtain different results when I use (even without nonlcon)

@(x) meancov(x,data)

Can you tell me how it works @(x)?

[Matt J]

You'll want to read the doc on Anonymous Functions, but basically they're a way of defining functions on-the-fly, possibly with fixed parameters, e.g.,

>>  fixed=2; f=@(x) x+fixed;  

>> f(10), f(20), f(30)

ans =

      12

ans =

      22

ans =

      32