How to use a sliced cell array as a reduction variable in parfor?

Hi I'm working on a code where I want to use cell arrays as a reduction variable.

You see, I have two matrices called (logErrOuter_1) and (logErrOuter_2) and based on them, I want to iteratively add those cells.

I cant because Matlab errors the valid indices. I wanna use parfor but my reduction variables are cell arrays.

%initialization
for j=1:3
    ZerrOuter_1{j} = zeros(j,1);
    XerrOuter_1{j} = zeros(j,1);
    ZerrOuter_2{j} = zeros(j,1);
    XerrOuter_2{j} = zeros(j,1);
end

%Outer syndrome
parfor i = 1:N
    %LogErrOuter
    logErrOuter_1 = func(...)
        logErrOuter_2 = func(...)
        
    %ZerrOuter/XerrOuter
    for j=1:3
        ZerrOuter_1{j} = ZerrOuter_1{j} + logErrOuter_1(logErrOuter_1(:,2) == j,3);
        XerrOuter_1{j} = XerrOuter_1{j} + logErrOuter_1(logErrOuter_1(:,2) == j,4);
        ZerrOuter_2{j} = ZerrOuter_2{j} + logErrOuter_2(logErrOuter_1(logErrOuter_1(:,2) == j,1),1);
        XerrOuter_2{j} = XerrOuter_2{j} + logErrOuter_2(logErrOuter_1(logErrOuter_1(:,2) == j,1),2);
    end
end

Answer 1

[Matt J]

The update terms like logErrOuter_1(logErrOuter_1(:,2) == j,3) need to be the same size in every iteration i in order to be compatible for accumulated addition. It's not clear why that would be the case.

But assuming you have a way to guarantee the sizes will always be the same, I think you would have to do something like below (which considers only one variable ZerrOuter_1 for simplicity). Then, after the loop, you can split ZerrOuter_1 up with mat2cell, and similarly with the other variables.

%initialization
ZerrOuter_1=0;

%Outer syndrome
parfor i = 1:N
    %LogErrOuter
    logErrOuter_1 = func(...)
        logErrOuter_2 = func(...)
        
    %ZerrOuter
    tmp_Zerr1=cell(3,1);
    for j=1:3
        tmp_Zerr1{j} = logErrOuter_1(logErrOuter_1(:,2) == j,3);
    end

    ZerrOuter_1 = ZerrOuter_1 + vertcat(tmp_Zerr1{:});
    
end

[mohadeseh]

Hi Matt, Yes the update terms have the same size in each iteration i, but different in j. and the problem showed up when I mixed sliced cell arrays with the update terms. your solution worked just as fine. Thanks a lot.