[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Next neural net question
From: |
Joern Thyssen |
Subject: |
Re: [Bug-gnubg] Next neural net question |
Date: |
Fri, 25 Jun 2004 13:27:49 +0000 |
User-agent: |
Mutt/1.4.2.1i |
On Fri, Jun 25, 2004 at 03:23:31PM +0200, Joachim Matussek wrote
> Hello,
>
> is it possible to use different neural nets with GnuBG? I don?t mean
> different weights but totally different nets, maybe another number of
> neurons or a different architecture.
Yes, depending on what you want to do it may be complicated.
The neural net files includes number of inputs, number of neurons in
hidden layer and number of outputs as well as all the weights and
threshold.
(1) you can change the number of neurons in the hidden layer just by
changing the file (and retraining of course)
(2) if you change number of inputs you have to modify the code in eval.c
as well
(3) if you want to add new neural nets you have to modify the code in
eval.c
(4) if you want to change number of outputs you need extensive changes
to eval.c (and possibly other files)
(5) if you want to use a totally different neural net evaluator this wil
also require extensive changes
(2) and (3) above is relative uncomplicated.
J?rn
pgpY8dJB5NSjW.pgp
Description: PGP signature