Re: An AI question and a UI question

Uzytkownik "Paul" <[EMAIL PROTECTED]> napisal w wiadomosci
news:[EMAIL PROTECTED]
> I can't figure out how to teach a computer what's a bad situation.
> Should I just do it the old-fashioned way?  I was thinking of having
> it determine what to do based on how close to death it is, with
> perfect health being 1 and dead being 0.  A .5 would be so-so, and a
> .01 would be pretty much worm food.  We can determine status through
> various things like HP, status effects like blindness, and maybe
> hunger.  I'm trying to think of some way of determining how well the
> creature is doing by some sort of formula.  Maybe Current HP/Max HP
*
> 1 if the creature can see, .5 if it's blind, and so on.  I don't
know
> what good values are for the weights, so I'd like a learning method
if
> possible.  Any ideas?

A general advice -- after completing a course on Neural Networks, and
a few practical implementations I came to the conclusion that there's
no real and worthy use of NN in roguelikes. The whole thing IMHO is a
big mistake, and NN is just one of those noisy "hurrah!.."s that
people make to impress other developers -- NNs are to slow, and the RL
environment is to complicated for any reasonable effect to be
achieved. Of course I may be wrong, but on your place, I'd rather
finish a game, and then think about adding a NN feature.

regards,
--
Kornel "Anubis" Kisielewicz
RLDev Code v.0.65
     L:FP E+ T+ R+++ P+ D++ G++ RL-- RLA+++ F:GearHead
     GenRogue Revolutions ( no website yet )
     W:DF Q+++ AI++ !GFX !SFX RN+++ PO--- Hp-- Re+++ S+++