[help] Error using ==> fmincon Too many input arguments

When I run a simulation of local constant cross validation for multivariables, I encountered the problem of too many input arguments.

function cv_m = lccvm(z,xc,xd,b,n) sum1 = 0 for i = 1:n dxc = (xc-xc(i,1))/(z(1)*n^(-1/5)) kc=exp(-0.5*dxc.^2); % continous kernel l=(xd==xd(i,1))+z(2)*(xd~=xd(i,1)); % discrete kernel k=kc.*l; % mixed kernel k(i,0)=0; % leave-one-out gx1=sum(b.*k)/sum(k); sum1=sum1+(b(i,1)-gx1)^2 end cv_m = 1/n*sum1;

and the bounds of z are z(1) from 0 to 20 z(2) from 0 to 1 there is no other constraint.

[Sarah Wait Zaranek]

Is lccvm your objective function for fmincon?

[Kang Wang]

yes it is. I also have a file for running loops.

[Kang Wang]

here is the fmincon file:

clear all;

close all;

n = 25;

loop = 100;

alpha = 1;

beta = 0.1;

gamma = 0.01;

for looop = 1:loop

xd = unidrnd(2,n,1)-1;

xc = randn(n,1);

u = randn(n,1);

y = alpha + beta*xd + gamma*xc + u;

z0 = [0.5;0.5];

lb = [0;0]

ub = [20;1]

[z(looop:1), cv(looop,1)]=fmincon(@lccvm,z0,lb,ub,@cons);

end

[Walter Roberson]

Refer to my Answer: it is exactly what is going on in your situation.

Answer 1

[Walter Roberson]

The objective function cannot accept that many variables directly. Please see the documentation of how to pass extra parameters.

[Kang Wang]

Thanks for your help. The problem is not solved yet. I tried to minimize the objective function by choosing z=[z1;z2]. When I tried to choose a scaler z = z to minimize the function, it worked by running [z_star(looop,1), lccv(looop,1)]=fminbnd(@(z) lccvm(z,xd,xc,y,n),0.1,2). Then I went back the original problem, I wrote code like the following:

(1)The cross validation file:

clear all;

close all;

n = 25;

loop = 100;

alpha = 1;

beta = 0.1;

gamma = 0.01;

for looop = 1:loop

xd = unidrnd(2,n,1)-1;

xc = randn(n,1);

u = randn(n,1);

y = alpha + beta*xd + gamma*xc + u;

z0 = [0.5;0.5];

lb = [0;0];

ub = [20;1];

lccv = @(z)lccvm(xc,xd,z,y,n); %Anonymous Function?

[z_star(looop,1), lccv(looop,1)]=fmincon(lccv,z0,[],[],[],[],lb,ub);

end

(2) the function file

function cv_m = lccvm(c,d,e,b,N)

sum1 = 0;

for i = 1:N

dxc = (c-c(i,1))/(e(1)*N^(-1/5));

kc=exp(-0.5*dxc.^2); % continous kernel

l=(d==d(i,1))+e(2)*(d~=d(i,1)); % discrete kernel

k=kc.*l; % mixed kernel

k(i,1)=0; % leave-one-out

gx1=sum(b.*k)/sum(k);

sum1=sum1+(b(i,1)-gx1)^2;

end

cv_m = 1/N*sum1;

I hope that you can show me the details about my problem. Sincerely.

Answer 2

[Ruslan Dautkhanov]

http://www.mathworks.com/matlabcentral/fileexchange/42770-logistic-regression-with-regularization-used-to-classify-hand-written-digits/content/Logistic%2520Regression%2520with%2520regularisation/fmincg.m

This one works better with too many arguments..