Sliced variable issue - variable cannot be classified

Dear All,

I'm trying to use parallel workers in an optimization problem but I cannot define X and Y as sliced variables. X contains the population, Y contains the fitness values. As the error message says: variable X & Y in a parfor cannot be classified.

First I define X, (it has to be outside the parfor loop otherwise I get new X-es in every single loop); than parfor follows:

parfor t=1:T    
    TmpZ=Func(Z(1,:));TmpX=Func(X(t,:));
    if TmpZ < TmpX
        X(t,:)=Z(1,:); Y(t,1)=TmpZ;
    else
        X(t,:)=X(t,:); Y(t,1)=TmpX;
    end
end

Thanks

Answer 1

[Edric Ellis]

I augmented your code a tiny bit to make it executable, and it works fine in MATLAB R2019a, and I checked all the way back to R2013b...

T = 7;
Func = @sum;
Z = rand(1, 3);
X = rand(T, 3);
parfor t=1:T    
    TmpZ=Func(Z(1,:));TmpX=Func(X(t,:));
    if TmpZ < TmpX
        X(t,:)=Z(1,:); Y(t,1)=TmpZ;
    else
        X(t,:)=X(t,:); Y(t,1)=TmpX;
    end
end

Answer 2

[Axel Groniewsky]

Dear Edric,

Thank you for dealing with my question. This part of the code runs fine yet, the error message says that 'The variable X in a parfor cannot be classified.' The other part of the code I have inside the parfor loop is:

R=randperm(T);
R(find(R==t))=[];
V=X(R(1),:)+F*(X(R(2),:)-X(R(3),:));

I do not alter X rather creating V. But you are right, without V it runs smoothly. Is there a way to define V in a parfor supported way?

Thanks!

[Edric Ellis]

This code is not order-independent, and cannot run inside parfor. Basically, the problem is that you're accessing elements of X that might or might not yet have been updated.

[Axel Groniewsky]

Another idea I tried is to save X after every iteration and load it before V vector is calculated. This way I end up having Transparency violation error. Is there another way you can think of which could dissolve this issue and let me use population-based metaheuristics with parfor? (I already tried the built-in GA but my own codes provide more accurate results.)

Thanks!