How to implement fminunc for a function with a variable lenght vector input?

I need to use fminunc to minimize a function that has some inputs that are constants, one input that's a vector which increases in length in each iteration and another input which is also an optimization variable.

So my original function would be something like this:

[loglike] = hyp(V, x, const1, const2)

Where V is a vector of variables that grows in each iteration V = [v1 v2 v3 ... vN] (for example at the third iteration V = [V1 V2 V3]) and x is the other optimization variable.

Going through the documentation I've found that fminunc needs all the inputs to be in a vector, so it's necessary to create an anonymous function:

f = @(w)hyp([w1 w2 w3 ... wN], wN+1, const1, const2)

And then use the fminunc function:

[x fval] = fminunc(f,w0)

However, I can't find a way to make this work.

[Matt J]

If the definition of your function is not fixed, but rather evolves "from iteration to iteration" as V grows, what does it mean to "minimize" it? Wouldn't its minimum also evolve?

[JoanOG]

Yes, the minimum changes in each iteration

[Matt J]

So, you're saying that you're trying to solve a sequence of minimization problems of different dimension?

[JoanOG]

It's a real time control algorithm, at each iteration the parameters of the Gaussian process have to be calculated using this minimization.

Answer 1

[Matt J]

Using vectorized commands, functions can be expressed in dimension-independent ways. For example, this function definition doesn't use any knowledge of length(unknowns):

 function fval=hypothetical(unknowns)

   x=unknowns(1);
   V=unknowns(2:end);

    fval=x.^3 +norm(V).^2;

 end

[JoanOG]

Got it working now! Thanks