Not setting a Random seed is no real problem. Without it, you still get random values. You just get the same every time you start your app. The problem here is that you don''t get a random number in the first place if it''s an interger randomizer.

Anyway, i read through the code again and I think I found the bug. And it''s not the randomizer. You are selecting the wrong individual after the roulette wheel selection. I worked it all out on paper with some numbers. The selection is fine (if Random returns a value between 0 and 1). But after this, you have this bit of code:

for k := 0 to Population[j - 1].NumberOfValues - 1 do
SelectionA[k] := Population[j - 1].Values[k];
AFitness := Population[j - 1].FitnessScore;

The individual that has been selected is j and not (j-1). So in the above bit of code change [j-1] to [j] and it should evolve a lot better.

Sander Maréchal
[Lone Wolves GD][RoboBlast][Articles][GD Emporium][Webdesign][E-mail]

Yes, I have taken that into account. Here''s a list of fitnesses for a population:

p[0]=10
p[1]=20
p[2]=50
p[3]=5
p[4]=15

TotalFitness=100

Now take a random number to select: Say 82. Your code should select p[3] here. (p[2] gets all values between 31 and 80. p[3] gets all values between 81 and 85). Lets follow your code:

SelectA=82-10=72; j=1; 72>10=true so continue
SelectA=72-20=52; j=2; 52>20=true so continue
SelectA=52-50=2; j=3; 2>50=false so STOP.

You have selected a population member now. j=3 and is correct. p[j] is the one you want. p[j-1]=p[2] and is incorrect (your code). I hope you can follow this. It makes a lot more sense working it al out on paper.



Sander Maréchal
[Lone Wolves GD][RoboBlast][Articles][GD Emporium][Webdesign][E-mail]

Nope, I haven't missed that. It just the way your selection code works. It is because you decrease the SelectA score first and then check against the given fitness after that. This means that you need to increment j by one to get the correct population member because you check against SelectA < fintess and not against SelectA < 0. After that, you need to decrease it again because you incremented one time too many in the loop (like you said). Total sum: j+1-1=j.

You could ofcourse rewrite your code to something like this (I don't know delphi but i'll copy-paste some)

{Select A}
RandomSelectionA := Random * TotalFitness;
j := 0;
repeat
begin
if (j > NumberOfPopulation - 1) then
j := 0;
RandomSelectionA := RandomSelectionA - Population[j].FitnessScore;
inc(j);
end;
until (RandomSelectionA < 0); {This is changed}

for k := 0 to Population[j - 1].NumberOfValues - 1 do
SelectionA[k] := Population[j - 1].Values[k];
AFitness := Population[j - 1].FitnessScore;

Makes sense. Say pop[0] has fitness 10 and SelectA=6. Thren it follows:
SelectA-pop[0].fitness; {SelectA=-4}
inc(j); {j=1}

SelectA<0 is true so you select j-1 (which is pop[0])



Sander Maréchal
[Lone Wolves GD][RoboBlast][Articles][GD Emporium][Webdesign][E-mail]

[edited by - smarechal on April 25, 2003 2:27:59 AM]