On 21/11/2008, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
>
> I can already create a ff net with
> net = newff([0 1; 0 1], [2 1], {"tansig","purelin"}, "trainlm", "learngdm",
> "mse");
>
> train it, use it, save it to file and load it back up. Now I like to (1)
> use the ga package to form a population of nets, (2) mutate them or one net
> of the population and (3) create offspring from two nets. And if that is all
> to easy, I like to rate them like e.g. using a function that adds together
> the wights within the net and the one with the highest is best.
>
> I saw that ga uses "NVARS" but haven't been able to figure out if that is
> one net, or the population and if the format is compatiple with what the
> feed forward nets use.
The simplest version of the function ga is
x = ga (fitnessfcn, nvars)
fitnessfcn must a function that, given a real vector of size
1-by-nvars, returns a real number.
This function returns the vector of size 1-by-nvars where fitnessfcnis minimum.
That is, you cannot give a network to the ga function.
You must find:
-a way to transform your net to a vector of real numbers
-a way to transform your problem to another problem: minimize a
function that, given a vector of real numbers, returns a real number
At the moment the ga package doesn't support
-constraints
-multi objective fitnessfcns
(that perhaps could be useful for your task).
At the moment I have not enough time for big new features in the ga
package, only for
-bugfixing
-reviewing patches
If you need a specific feature of the MATLAB counterpart of the ga
function, or you want to propose a patch, please let me now (on this
list).
Hope this helps,
Luca Favatella
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Octave-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/octave-dev
