Author: José Carlos
Date: 14:18:31 03/23/02
Go up one level in this thread
On March 23, 2002 at 15:47:34, Marc van Hal wrote: >I just thought aslong as computer programs don't have the positional knowledge >For all positions. >It might be smart to first use an normal preferably big book. >Then play many games with this book. >And save all games. >Then look in your database in which openings your programs does best >And make this the mainlines. > >(In fact I have done this before,with great sucses.) >Keep on playing many games. >Then you build a database put in all the games of your program >Build a book from it. >Then you look which games where lost by your program and lock these moves. >you even could go further with also giving the positions which where drawn by >your program a lower valeu. >(That's below 50% or that 50% move are seldom played moves.) >The idea I did get because if you have a patser program it is always bad to get >an opening which results in a quick endgame. >For a positional program it is the otherway around it prefers endgames! >And it is a also sort of pre knowledge which has been used. > >It also makes it posible if your program is not good in all ellements >opening,midle and endgame >positional and tactical play . >it still can get a nice score. >For Chrisophe a special openings book for Tiger15 > >And a special openings book for Gambit Tiger15 >Could be an example > >Ps Christophe this one is for free >My Alexander setting are not the transpose from midle to endgame anymore. >Still the settings are logic. > >Regards Marc van Hal Some time ago, I thought exactly the same as you say, but this takes too much time for a person, so I implemented book learning in Averno. My book learning, on the one hand, extends the book lines; on the other hand, gives plus points to lines where the program wins and minus points to the lines where the program loses. Then, when selecting moves in a game, the program knows what lines are best for it, and tries to play them over and over. It can even invent new moves if a line looks promising but at some point, all the book moves look bad. As you say, this improves the results by a big margin. 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.