using lsqlin when you have a large matrix

hi,

Can someone please tell me how to correct the following code.

x = [yt,r];
a = eye(2);
b = zeros(2,1);

opts = optimset('lsqlin');
opts.LargeScale = 'off';
opts.Display = 'none';
coef = lsqlin(x,yt2,-a,-b);

My x is a 1000*2 matrix. so i get an error. I read the help for that but couldn't understand it clearly.

thanks.

Answer 1

[Teja Muppirala]

I don't think there is anything wrong with this code, except LSQNONNEG would be easier.

coef = lsqlin(x,yt2,-a,-b);

is the same as

coef = lsqnonneg(x,yt2);

When I run your core, I don't get any errors at all. I just get

coef =

    0.4036
         0

Which I have no reason to doubt is a valid answer.

Are you perhaps referring to the warning that comes up regarding the solver?

[dav]

thanks I also figured it out that coef = lsqnonneg(x,yt2) is much easier.

Answer 2

[Matt J]

If you have 1000 unknowns why is "a" only 2x2?

Also, since your constraints are only lower bounds on the unknowns, use the lb input argument instead of A,b

Also, since you only have positivity constraints, consider using LSQNONNEG instead of LSQLIN.

[dav]

thanks. I am relatively new to matlab. My code is given below. is there any way to correct that code?

a0 = 0.05; a1 = 0.1; b1 = 0.85;
nu = randn(2300,1);
epsi = zeros(2300,1);
h = zeros(2300,1);

for i=2: 2300 
    h(i) = a0 + a1 * epsi(i-1)^2 + b1 * h(i-1)  ; 
    epsi(i) = nu(i) * sqrt(h(i));

end

yt = zeros(2300,1);

for i=1: 2300 
    yt(i) = epsi(i)*epsi(i);
end

order = 5;
m = arx(yt, order);

r = resid([yt(1:order);yt], m);
r = r(order+1:end);

yt2 = zeros(2300,1);

for i=1: 2300 
    yt2(i) = yt(i)- r(i);
end

yt2(1:1300) = [];

yt(1:1299)=[];
yt(1001:1001)=[];
r(1:1299)=[];
r(1001:1001)=[];

x = [yt,r];
a = eye(2);
b = zeros(2,1);

opts = optimset('lsqlin');
opts.LargeScale = 'off';
opts.Display = 'none';
coef = lsqlin(x,yt2,-a,-b);

[Walter Roberson]

Even if you only have two constraints that you want to actively make use of, you need to supply values for the others, possibly "inf".

[Matt J]

is there any way to correct that code?

As I already mentioned, it is an error to use a 2x2 matrix a=eye(2) and 2x1 vector b=zeros(2,1) when you have more than 2 unknowns.

[Teja Muppirala]

There are not 1000 unknowns, there are only 2. It only seems that way because what the documentation refers to as A, is being called x here, and what the doc refers to as x, is being called coef.

[Teja Muppirala]

Oops I meant C, not A.

[Matt J]

There are not 1000 unknowns, there are only 2.

Ah well. The question has been edited.