Author: Fernando Villegas
Date: 16:44:59 10/22/00
Go up one level in this thread
On October 22, 2000 at 17:13:16, Uri Blass wrote: >On October 22, 2000 at 17:06:17, Ratko V Tomic wrote: > >>> 1)I believe that the difference between Gambittiger and the default version of >>> tiger is the fact that Gambit has bigger positional scores. >> >>That wouldn't account for having two separate programs and the intent Christophe >>mentioned to unify the two in the future and get the best of both worlds. If it >>were just evaluation terms and weights one could just have a single program with >>changed settings (using the union of all the evaluation terms in the two Tigers, >>then if the setting "i" has weight(i)=0 don't evaluate term T(i)). It must be a >>greater difference than just tweaking the weights of various evaluation terms. >>And if were that simple everyone would tweak their's, too, but apparently that >>alone doesn't work very well (as Bob can tell you). > >I do not say that the only difference is numbers but my understanding from >previous posts is that the difference is simply the evaluation function and not >new search rules. > >I did not say that it is a simple to do it. >I only said that I do not call it a new paradigm. > >Uri Hi: Maybe the word "paradigm" is too big and ambitious and maybe the changes that empirically were made in the original Tiger code in order for this last to become Gambit Tiger were few and restricted; even more, maybe they were not more than what you say, higher evaluation to squares arround the enemy king or so, BUT even if this is so, it amounts to a different kind of programming. I think there is great difference between launching an attack to h7, even with sacrifice, if the program already calculated a clear winning with that, to launchingg an attack with the very same moves as in the first case IF this course of action was chosen because of the general reason it put the enemy king on the open. If you look at it from outside you could even believe it is the same program because it producec the same output, BUT it would be a very different animal. As in human chess many very different players can play the same lines for a long while BUT on the ground of oposed reasons, so more than a program maybe could replay many moves by Gambit being, anyway, an absolutely different entity. In the case of evaluating highly some squares just for positional reasons, I agree with you that we have there "just" that, a positional evaluation, but I think that such an evaluation, if it is made inside the normal frame of thinking of a normal proggram, has nothing in common with the one that maybe Gambit perform. In the first case the high score just is used to call the attention of the program to see at it, just in case there is something, but the last word about that "somethingg" will be of the normal evaluation function in concrete terms; in the second case the evaluation could be considered as valuable just in itself. Of course this is not so balnck and white. Probably we are facing a nuanced issue with a too much clumsy wording. Fernando
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.