The variable newMap in a parfor cannot be classified.

I am trying to store container map with fos value (returns from fun_foo function using x as key and fos as value) and if x already called before map will retuen correct value and I can reduce call of fun_foo function.

Goal is to make this program as fast as possible.

newMap= containers.Map();
parfor i=1:100     
    tf = isKey(newMap, mat2str(x(i,:)));          
    if tf == 1
        fos(i,1) = newMap(mat2str(x(i,:)));
        gg = gg+1;
    else
        fos(i,1) = fun_foo(x(i,:));
        newMap(mat2str(x(i,:))) = fos(i,1);
    end
end

Answer 1

[Edric Ellis]

This loop cannot run because the variable newMap is being read from and written to in arbitrary locations as the loop progresses. This is order-dependent behaviour, and parfor disallows that.

So far as I can tell, you're using newMap as a cache to avoid recomputing fun_foo for rows of x that you've already seen. It might therefore work to ensure that you run your parfor loop over only the unique rows of x. Something like this:

% Make some random data - this will probably
% have duplicated rows.
x = randi(5, 100, 3);
% Dummy function for testing
fun_foo = @sum;

% Extract the unique rows, keep the "mapping" vector
% ic for later
[xu, ~, ic] = unique(x, 'rows');
parfor i=1:size(xu, 1)
    % We know that each row is unique, and hasn't
    % been computed before, so just do:
    fos_u(i,1) = fun_foo(xu(i,:));
end
% To get the original "fos", we need to undo the ordering
% from the call to 'unique'
fos = fos_u(ic);