inner matrix dimensions must agree error in the function?
Show older comments
function [f] = objfun(W,VCVMatx)
%function for varience of the portfolio with swtN stocks
f =W*VCVMatx*W';
this is my function I am trying to run in the main program. However, when run in the main program, it keeps giving me :
Error using * Inner matrix dimensions must agree.
W is a 10(N)X1 vector, while VCVMatx is an NXN vector?
Does anyone see anything wrong, considering I multiplied it by a transposed W'
I tried running fmincon, but in the main program it says the function matrix dimensions do not agree.
f = (W'*VCVMatx)*W;
i=1;
while i<=rows(ERGrid);
ExpectedReturn = ERInc(i);
[W] = fmincon('objfun',x0,swtOther_A,swtOther_b,swtOther_Aeq,swtOther_beq,swtMinWt1,swtMaxWt1,[],options);
ERGrid(2:swtN+1,i) = W;
i=i+1;
end;
W' is a 1X10 vector while VCVMatx is a 10X10 matrix and W is 10X1 vector
fmincon cannot run as a result.
Answers (1)
Geoff Hayes
on 30 May 2014
The first matrix product (in the equation) is the problem. Since W is Nx1 and VCVMatx is NxN, then W*VCVMtx will fail due to the incompatible dimensions of the two matrices. If you are just expecting a single value, then the equation should be re-written as
f = W'*VCVMtx*W;
29 Comments
Sameer
on 30 May 2014
Star Strider
on 30 May 2014
If you type:
W_size = size(W)
VCVMatx_size(VCVMatx)
what are the results?
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
Type the commands that Star Strider indicated just prior to evaluating f to ensure that the dimensions of the matrices are compatible.
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
Edited: Geoff Hayes
on 30 May 2014
Try the following to get the number of rows and columns of VCVMatx
VCVMatx_size = size(VCVMatx)
Sameer
on 30 May 2014
Star Strider
on 30 May 2014
I’ve never seen the size function return ‘undefined’ as output.
Go back and see how you created VCVMtx. If the code creating it is not very long, post it or the relevant part of it.
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
@Sameer - you said a couple of comments ago that W is 1x10 (which is different from the original statement of 10x1). If that is the case, and VCVMtx is still 10x10 (what did size(VCVMtx) return?), then your original line of code (not the one I suggested above) should be valid
f = W*VCV*W';
Sameer
on 30 May 2014
Star Strider
on 30 May 2014
What is:
swtN_size = size(swtN)
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
What then is:
VCVMatx_size = size(VCVMatx)
Sameer
on 30 May 2014
Sameer
on 30 May 2014
Star Strider
on 30 May 2014
So:
f = W'*VCV*W;
should be a scalar.
Geoff Hayes
on 30 May 2014
Strange that it doesn't work in the function. Could you attach the function (m file) to a comment, using the paper clip button?
Sameer
on 30 May 2014
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
Edited: Geoff Hayes
on 30 May 2014
The code is
function [f] = objfun(W,VCVMatx)
%function for varience of the portfolio with swtN stocks
global VCVMatx;
W_size = size(W);
VCVMatx_size = size(VCVMatx);
f = (W'*VCVMatx)*W;
end
Why is there a global variable with the same name as the input parameter? If this global variable has not been set then it is probably empty and so the error is thrown. Please comment out the line
% global VXVMatx;
and try again so that the second input (to the function) is used.
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
Okay - I don't have the fmincon function, but reading the documentation at http://www.mathworks.com/help/optim/ug/fmincon.html indicates that the function handle passed into fmincon must be for a function that accepts a vector and returns a scalar. Is that why you have tried to get around the limitation of only one input being passed - by using a global variable?
If that is the case, then I suppose you have to put that global variable back and remove the second input to your objfun. Then, assuming that the input vector is 1x10 and that the global variable VCVMtx is 10x10, then your f should be changed to
f = (W*VCVMatx)*W';
and the global variable VCVMatx must be set prior to the evaluation of fmincon.
Sameer
on 30 May 2014
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
I think the problem that you were originally experiencing was the the global variable VCVMatx was not defined to be a 10x10 matrix, it was just empty. And so if you multiply a non-empty matrix ( W ) with an empty matrix ( VCVMatx ) the message Error using * Inner matrix dimensions must agree is thrown. So it didn't matter which combination of f we were using (transpose of W on left or right) it would still fail.
At some point, the global variable must have been set. And now, with the correct order of the matrix multiplications for f, it works as expected.
But you will have to remember to set the global variable VCVMatx...
Sameer
on 30 May 2014
Geoff Hayes
on 30 May 2014
No, unfortunately I have never used that function (nor have the toolbox for it).
Categories
Find more on Creating and Concatenating Matrices in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!