Author: José Carlos
Date: 02:59:04 08/17/01
Go up one level in this thread
On August 16, 2001 at 15:48:22, Marc van Hal wrote: >Many openingsbooks used today look like a travel back in time. > >Which is most of the time funy. > >But also an other question pops up Is the Chessmasterbook still bad now. > >I gues it gets more changes then it did get 2-3 years ago. I'm writing the book for my new program now, and your post raises an interesting question: how to create a good (for a certain program) book? It seems a simple question, but really it hides a lot of details. First, it's difficult and takes too much time not to use some sort of automated creation from a set of games. I've written my code so that I can easyly add handly variriations, and disable the use of some moves with the '?' sign, or encourage others with a '!'. Additionally, I use different files for black and white positions. From a set of top GM games, I chose 1-0 or 1/2 for the white file and 0-1 or 1/2 for black. Learning ability so that it avoids variations where it loses often. And hand tuned variations where I tell the program, for example "the French is bad for black, don't play it". Pretty standard stuff so far. This leads to acceptable books but, how to create _really good_ books? For example, if you know your program is a good attacker, include gambits and sharp positions in the book. This is very obvious. I've tried to figure out how to provide something new... and failed. The main problem I see in automated learning is that needs slow games to be reliable and then it is too slow. And if you upgrade your program, the learning from previous versions is obsolete. Any ideas out there? José C.
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.