Error using fminunc with array as function output and multiple inputs

We are trying to optimize three parameters using fminunc.

Our function contains two for-loops and gives us a probability over time in a vector.

Our parameters for optimization are used in this function as x(1), x(2), x(3)

function P = probabilityFunction(x,T,Tmed)
    
    P = ones(length(T),1)*x(3); 
    
    for t=0:T(end)
        numPills = sum(Tmed < t);
        for i=1:numPills
            P(t) = P(t) - x(1)*exp(-x(2)*(t-Tmed(i)));
            
        end
    end
end

When we try to run the fminunc, it gives us the error "Supplied objective function must return a scalar value"

We call the fminunc like this:

% pooling the data into one array and sorting
Tmed = [deltaT_mo deltaT_ev]';
Tmed = sort(Tmed); 
T = (1:max(Tmed))'; 
noDrug = 0.5; 
E = 0.0025; 
v = 0.009; 
x0 = [E; v; noDrug];

%%
f = @(x)probabilityFunction(x,T,Tmed); 
%%
[x,fval] = fminunc(f,x0);

Should we use another function or is there a way to make the probability as a scalar?

- Thanks

Answer 1

[Alan Weiss]

It is not clear what you are trying to do. Your P(x,T,Tmed) is not a scalar, so it is not clear what it means to minimize it.

You might be looking for a least-squares minimization, something like sum(P.^2). If so, you can take your existing objective function and ask lsqnonlin to minimize it. Or you can reformulate your objective function as the sum of squares. For details, see Nonlinear Data-Fitting.

Alan Weiss

MATLAB mathematical toolbox documentation