Author: Robert Hyatt
Date: 07:57:11 08/16/02
Go up one level in this thread
On August 16, 2002 at 09:50:10, Pham Hong Nguyen wrote: >>The book you are using is _intentionally_ very large and wide, for use on >>chess servers. Narrow books get killed if you play thousands of games >>with them, because they don't have much variability. >> >>Book learning is the thing that ultimately solves this problem for >>automatically-generated books... That was why it was developed... > > >I am wondering why you didn't update book from data of learning? You should have >a huge and excellent data. > >I am planing update my book after every 1000 games, then I will have a new and >better book without old data of learning :) >PHN I am not quite sure what you mean. But there _is_ a danger to long-term book learning. If the program gets better over time, then lines it used to do poorly in might become winning lines now. Learning hides that. Lines that used to be good due to nothing more than luck, may become bad because a move that now looks to be more correct on general principles is really worse over the long-term.
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.