Author: José Carlos
Date: 00:44:55 01/30/03
Go up one level in this thread
On January 29, 2003 at 14:47:05, Dann Corbit wrote: >On January 29, 2003 at 03:08:48, José Carlos wrote: >[snip] >> The idea is good and I plan to do something similar if someday my program is >>strong enough to trust such searches. But realize that every position in a book >>is an exit position, so you need to analyze all of them. >> I prefer the idea of verifying book during games and then dump that into the >>learning file so that it won't be needed anymore. IMO, learning is the way to >>build a solid book, but not the aggresive learning some programs do in order to >>win a match, but a more long-term learning. > >A good way to achieve it is with statistics. >Store lots of information about book positions. >Wins/Losses/Draws/Unfinished >MCO rating >NCO rating >ECO rating >Computer centipawn evaluation and depth (for several engines) >Undeperformance/Overperformance (IOW -- when this move is played, is the winning >expectancy in line with ELO figures or does it do better than ELO would >estimate?) >SuperGM preference >GM preference >IM preference > >Probably lots more could be stored. Maybe in a real database. Yes, that's good information. I did part of that in Averno, the part related to the engine, this is: save eval and depth, along with final result of the game and small recalculation of previous moves according to search results and game result. 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.