Author: scott farrell
Date: 06:45:19 11/28/02
Go up one level in this thread
On November 28, 2002 at 04:23:15, Sune Fischer wrote: >My present book format is really simple, too simple. > >I would like to use transpositions and book learning, so I need a new format. > >Is there a "standard" way of doing this, some well known technique which >includes statistics for each line and makes it possible to transpose back into >the book? > >I can only think of primitive ideas, like storing the full FEN and append each >possible move reply to it. That will of course blow the book to a huge file. I >guess some other hashing scheme can be done, but what is the _best_ way? :) > >-S. I have taken a different approach to most. I actually run a largish relational database (mysql). The db is a couple of hundred mbs, and the main index if about 60Mb, which I tell mysql to hold in RAM. The db reacts pretty quickly during a game. They keys are all based on zorbist keys, so I get transpositions etc. I am a commercial programmer, so database coding is too easy, and doesnt require too much thought. The idea being I can implement whatever sort of stats and learning etc, easily, without too much coding. I have about 40 columns of data for each position of about 6 million positions in my opening book. I have the win/loss/draw and avg rating of each win/loss/draw, and have 2 brackets of win/loss/draw stats, 1 being those players where myrating>oprating-200, and then obviously the opposite of that. Then I do a stupedenous amount of stats to select the move, my MYSQL does all that for me ..... I also use SCID to manage a db games, and use SCID to query out the required PGN, and have an import routine to read the SCID output PGN. I strongly suggest you use SCID to query your game PGNs. Scott
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.