Unable to perform assignment because the size of the left side is 1-by-1 and the size of the right side is 1-by-712.

Hi everyone,

I am trying to fit some data using "fminsearch" but i get the following problem "Unable to perform assignment because the size of the left side is 1-by-1 and the size of the right side is 1-by-712."

Here are the four lines.

var = aSlipFA(cornPOS(1,1):cornPOS(1,2));

ErrorMagicFormula =@(c) c(3)*sin(c(2)*atan(c(1)*var - c(4)*(c(1)*var - atan(c(1)*var)))) - FLatFA(cornPOS(1,1):cornPOS(1,2));

c0 = [2000,1.4,max(MyFLatFA(cornPOS(1,1):cornPOS(1,2))),0.9];

fit1 = fminsearch(ErrorMagicFormula,c0);

Thank you very much!

Answer 1

[Raunak Gupta]

Hi,

In my understanding you are trying to optimize the ‘ErrorMagicFormula’ using fminsearch. The “Unable to Perform assignment” error message is due to the vector that is returned from the ‘ErrorMagicFormula’ calculation. The input to fminsearch should be a function that return scalar, so it is recommended to change the ‘ErrorMagicFormula’ such that it returns a scalar by using some transformation.

Here the variables var and FLatFA(cornPOS(1,1):cornPOS(1,2)) is supposedly a 1-by-712 vector.

[Massimiliano Vecoli]

Hi, Thank you for your answer !

I am trying to minimize ErrorMagicFormula so that the first part of this function approximates FLatFA(cornPOS(1,1):cornPOS(1,2)) at best. My goal was to make a function which had the variable "var" already defined, but with some open parameters to be optimized. Do you think this is possible ?

Thank you again

[Walter Roberson]

Do you think this is possible

No. If you are not willing to reformulate to calculate a scalar, then the best you can do is use gamultiobj() to get pareto fronts.

You should consider the possibility of calculating sum-of-squared-residues,

sum( (projected_values - actual_values).^2 )

In other words, a least-squared fitting. This gives a scalar result that can be minimized (but you will not necessarily find the global minima.)

[Raunak Gupta]

The main concern is that the ErrorMagicFormula should return a scalar because whenever we do optimization the function that need to be optimized should be a number. If you are getting lets say a 1-by-712 vector then you may transform it by vector multiplication with a 712-by-1 vector thus resulting into a scalar value at the end. Thus this function can be minimized using fminsearch.

% Lets assume f is returned from ErrorMagicFormula which has size 1-by-712
f = randn(1,712);
% taking a coefficient vector randomly
coeff = rand(712,1);
final_function = f*coeff;
% below can be the function handle for ErrorMagicFormula
% ErrorMagicFormula = @(c) (c(3)*sin(c(2)*atan(c(1)*var - c(4)*(c(1)*var - atan(c(1)*var)))) - FLatFA(cornPOS(1,1):cornPOS(1,2)))*coeff;

[Walter Roberson]

Small ammendment: if you multiply a 1 x 712 and a 712 x 1 and expect a 1 x 1 result, that is not vector multiplication, that is matrix multiplication. vector multiplication would be the .* operator and would give you a 712 x 712 result.

[Raunak Gupta]

Thanks, for the clarification.