deterministic array filling in parfor loops

this code:

fit = [];
parfor i = 1:10
    pause(3*rand());
    fit = [fit; i];
    disp(i);
end

gives every parallel worker a different end time. disp(i) shows an expectedly nondeterministic number output. eg.:

 2, 1, 7, 6, 4, 3, 5, 8, 9,10

now, if we look at the array 'fit', which I would have expeced to be filled with a series of numbers equal to the output, it shows as:

1,2,3,4,5,6,7,8,9,10

Is this expected behavior? How do the workers know what position of the array they should write to?

best wishes!

Answer 1

[Matt J]

PARFOR pre-categorizes 'fit' as a reduction variable before the loop is run. Essentially, it knows in advance that

fit = [fit; i];

can be processed like,

parfor i=1:10

   fit{i}= whatever

end

fit = cell2mat(fit)