Matrix variables cannot be classified in PARFOR loop in Matlab

Question

In my for loop, I set a matrix variable ga to save the results of every loop. But when I change the for loop to the parfor loop (to accelerate), there is a warning as follows:

And when I run the code, I get another error:

The code is:

R=100;
alpha_set = [1,2,3,4,5]; % This is an index set
ga = zeros(2,5); % to save results of addition

parfor h=1:R

[A1,A2] = random_sample(A,0.6);

...

for ai=1:5
    alpha = alpha_set(ai);
    ga(1,ai) = ga(1,ai) + T_lower(A2,alpha)/R;
    ga(2,ai) = ga(2,ai) + T_upper(A2,alpha)/R; % accumulation
end

end

T_upper and T_lower are both functions which return numbers.

I want to sum the returns of the two functions, and save the values under different index alpha to different positions of ga, so ga should be classified as reduced variable, shouldn't it? (While Matlab cannot classify it.)

How can I debug the code and make parfor run successfully?

H Wheelwright · Accepted Answer · 2018-08-20 07:53:08Z

1

You define the array to hold the result before the parfor loop, but then try and access it from within the parallel loop. As noted, Matlab cannot classify the variable. The problem is that ga is indexed inside the nested for-loop. The code below uses a variable that is indexed differently to address this issue.

R=100;
alpha_set = [1,2,3,4,5]; % This is an index set 
N = 2;
gas = zeros(R,length(alpha_set),N);



parfor h=1:R

    A = 1.0;
    [A1,A2] = random_sample(A,0.6);


    for ai=1:5
        alpha = alpha_set(ai);
        for ni = 1:N
            switch ni
                case 1
                    gas(h,ai,ni) = T_lower(A2,alpha)/R;
                case 2
                    gas(h,ai,ni) = T_upper(A2,alpha)/R;
             end
        end
    end


end

gaResults = zeros(N,length(alpha_set));

for ni = 1:N 
    gaResults(ni,:) = sum(gas(:,:,ni),1);
end

function [ output ] = T_lower( a1,a2 )
output = a1*a2;
end

function [ output ] = T_upper( a1,a2 )
output = a1+a2;
end

function [o1,o2] = random_sample(a1,a2) 
output = a1 + a2.*randn(1,2);
o1 = output(1);
o2 = output(2);
end

edited Aug 20, 2018 at 7:53

answered Aug 17, 2018 at 9:37

H Wheelwright

338 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Ma Xingkong Over a year ago

However, when I have to use a matrix to store the results of different functions with a similar index, this method is tedious. For example, if I have a family of functions, such as T_1,T_2, ...,T_n, then I have to define ga1, ga2, ..., ga_n. It'll be a frustrating repetition. How to avoid this problem?

H Wheelwright Over a year ago

Perhaps you could increase the dimensions of the results array? For example see the edited answer.

Collectives™ on Stack Overflow

Matrix variables cannot be classified in PARFOR loop in Matlab

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related