I know this might be a weird question, but is there a standard way of naming Chromosomes of a NN? I''ve seen a few programs that did this and I was wondering what the convention would be if there was one. Anybody?

You''re mixing metaphors--your brain isn''t defined in chomosomes. Just use names that are clear. In all the time I''ve used neural networks I haven''t seen any standard scheme, although I sometimes name layers things like cerebellum and brainstem but that''s more me being a smartass as opposed to doing anything to make things clearer.

Ok perhaps I may have meant genotype. But I''m not looking for what you call it. I am trying to see what others would label their creatures based on the weights and values they hold. In other words, lets say that you were to find a way of graphing and numbering each neuron and weight so that you can examine and compare the values of each and label/catagory species or sub-species based on that. I''m probably being a bit creative with the concept. But giving in to this idea, what would you use as a convention?

what are you talking about, Neural Networks or Genetic Algorithms? Because they are seperate entities.

In short I''m talking about the guts of a NN.

A nueral network created by a genetic algorithm? :D

Yes. This is the only way I know of evolving a population of NNs. I figured this was safe to assume.

[edited by - WILL at RedAnt on April 13, 2003 1:35:48 AM]

wow, that''s some pretty advanced stuff, using a GA to create a neural net.
Basically, AI programming, especially involving NN and GA, is so young that it hasn''t created any conventions yet (please, someone contradict me).
Just remember, use names that are short enough to type, but descriptive enough to not require a comment. As long as you do that, you should have a perfectly fine name for you variables regardless of any convention.
Conventions are really only good on huge projects with many programmers, because each programmer needs to be able to predict what variable names the others will use.
For me, naming my chromosomes has been a moot point. I''ve always stored my chromes in an array to help facilitate crossover while breeding, and I always gave the array a name like nChromo (n to indicate it''s an array of integers in a sort of Hungarian Notation).
As to the purpose of each individual chrom, I made a reference table, put it in a promonent position in my code, and wrote it on a piece of paper. I think this is the best way, considering real DNA doesn''t have little labels telling us what each sequence does.

AI junkie uses GA to evolve Neural Nets. It uses a vector of doubles as a chromosome and each double a weight in the Neural Net. Also, there''s a new tutorial on AI depot which uses GA not only to evolve the weights of the ANN but the structure of the ANN as well. This GA can add, remove etc, nodes and connections from a neural net thus evolving an optimal ANN achitecture with optimal weights.

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

this has been done numerous times in numerous ways. I have seen a few articles about it. Look at citeseer (dont remember the URL but look in google). it should give a few articles about the subject. If i remember correctly, this approach was used to make a GO player....

--Spencer

--Spencer

"Relax, this dragon is sleeping..."