gsoleilhac / nsgaii.jl Goto Github PK

I'm attempting to execute the example code (knapsack), but I get this error:

ERROR: LoadError: UndefVarError: undef not defined
Stacktrace:
[1] (::Atom.##110#114{String,String})() at C:\Users\leona\AppData\Local\JuliaPro-0.6.2.2\pkgs-0.6.2
.2\v0.6\Atom\src\eval.jl:104
[2] withpath(::Atom.##110#114{String,String}, ::String) at C:\\\\\\\\\\\\\\\\\\\
\\\\\\\\\Users\leona\AppData\Local\JuliaPro-0.6.2.2\pkgs-0.6.2.2\v0.6\CodeTools\src\utils.jl
:30
[3] withpath(::Function, ::String) at C:\Users\leona\AppData\Local\JuliaPro-0.6.2.2\pkgs-0.6.2.2\v0
.6\Atom\src\eval.jl:38
[4] hideprompt(::Atom.##109#113{String,String}) at C:\Users\leona\AppData\Local\JuliaPro-0.6.2.2\pk
gs-0.6.2.2\v0.6\Atom\src\repl.jl:65
[5] macro expansion at C:\Users\leona\AppData\Local\JuliaPro-0.6.2.2\pkgs-0.6.2.2\v0.6\Atom\src\eva
l.jl:99 [inlined]
[6] (::Atom.##108#112{Dict{String,Any}})() at .\task.jl:80
while loading C:\Users\leona\Downloads\knapsack_test.jl, in expression starting on line 13

Checking the source code, the problem seems to be in 'functions.jl' line 5 where you write undef, which isn't indeed declared anywhere. What's that? Is it a way to say 'not implemented yet'? Am I missing something?

With some updates, the readme doesn't quite work as written. Specifically, the calls to bitrand need to be changed, and you need to call using Random, LinearAlgebra

The following worked for me:
`using NSGAII
using Random
using LinearAlgebra
n = 20 #Number of items
p1 = [77,94,71,63,96,82,85,75,72,91,99,63,84,87,79,94,90,60,69,62] #Coeffs objective 1
p2 = [65,90,90,77,95,84,70,94,66,92,74,97,60,60,65,97,93,60,69,74] #Coeffs objective 2
w = [80,87,68,72,66,77,99,85,70,93,98,72,100,89,67,86,91,79,71,99] #Items weights
c = 900 #Knapsack capacity

rng = MersenneTwister(1234);
popsize = 100
nbgen = 200

init() = bitrand(rng, n) #our genotype is a binary vector of size n, initialized randomly
z(x) = dot(x, p1), dot(x, p2) #and our objectives are the sum of the items we pick

function CV(x)
sumW = dot(x, w)
return sumW <= c ? 0 : sumW - c
end

#We can now call
result = nsga_max(popsize, nbgen, z, init, fCV = CV)
`

The model I am using means that the cost of evaluating genes is very high. Can there be an argument to nsga that allows you to use pmap for the parts of nsga that involve computing new genes? If I try to implement this on my fork which parts of NSGAII.jl should I edit?

Hello, while inspecting your code I noticed a potential flaw in the crowding distance assignment function.
I see that you split crowding_distance_assignement! in two cases: one for when you have only 2 objective functions, the other for more than 2.
I assume this is an optimization to avoid a sort in the first case, which is fine and makes sense.
However, I think that you forgot to check if pop[1].y[1] != pop[end].y[1] and if pop[1].y[2] != pop[end].y[2] in the first case, which could become a source of zero division errors when the whole population shares the same fitness for at least one objective.
In the second case, instead, you perform the check (line 112) and that's good.
I think it's just an oversight, because I see no reason to assume pop[end].y[1] always different from pop[1].y[1]

PS: I also suggest to rename 'assignement' into 'assignment'