Parallel the evaluation step of a genetic algorithm

New to Julia
1

What is the best way to parallel my evaluation step of the genetic algorithm? I.e. would like to parallel the following code.

function test_func(pop::Pop, extra_data::Dict)::Pop
    for i in 1:pop.num_size
        pop.obj_func[i, :f1] = (4*pop.individuals[i].parms[1][1]*pop.individuals[i].parms[1][1])+(4*pop.individuals[i].parms[2][1]*pop.individuals[i].parms[2][1])
        pop.obj_func[i, :f2] = ((pop.individuals[i].parms[1][1]-5)*(pop.individuals[i].parms[1][1]-5))+((pop.individuals[i].parms[2][1]-5)*(pop.individuals[i].parms[2][1]-5))
    end
    return pop
end

mutable struct Pop
	num_size::Int64
	individuals::Array{Ind,1}
	obj_func::DataFrame
end

mutable struct Ind
    parms::Array{Array{Number,2},1}
end
2

Adding a Threads.@threads in front of the for loop should be the easiest way.

3

I already have done it and it didn´t lead to an error, but I´m not sure if it is thread safe, because I read in some topics that not all types are thread safe.

4

Yeah, it seems a little unsafe because you can have multiple threads trying to write pop.obj_func, that might lead to complications.

5

How do I know that it works correctly or not? Just by comparing the result of the parallel executed loop with the serial executed loop? Can I just store the data first in a simple array and than copy it after the for-loop to the dataframe pop.obj_func. Have I also consider the reading part? Let´s say my evaluation function is a bit more complicated. I.e. I have a mutable struct, which is a temporary variable, changes in every loop, is depended on the pop.individuals[i].parms[1] and necessary to calculate the objective functions.

function test_func(pop::Pop, extra_data::Dict)::Pop
    for i in 1:pop.num_size
        my_struct = MyStruct.create(pop.individuals[i].parms[1], extra_data)

        pop.obj_func[i, :f1] = calculate_f1(my_struct)
        pop.obj_func[i, :f2] = calculate_f2(my_struct)
    end
    return pop
end
6

I tried it out and compared the serial executed loop and the parallel executed loop. It seems to be thread safe, because I got the same results.