Author: Dave Gomboc
Date: 21:24:08 08/22/99
Go up one level in this thread
On August 22, 1999 at 21:49:30, Jerry Creed wrote: >First, what are your recommendations for the settings of the Zarkov 4.0 >engine. I'm using +- 100 or 1 pawn, +/- at 50 or a half a pawn and += at >25 or quarter pawn. How does that sit with everyone? My experience is that these values are a little too tight. I use 0.35, 0.83, and 1.30 pawns for these categories. >Suppose I want to study my games as white. I make aup a pgn file in >Chessbase and create a new book in Bookup and import those pgn games. If I >backsolve, I get all kinds of weird assessments, for instance - for d4, - >for c4, - for Nc3 because I eventually lost the game by some move up near 20 >or something. But there isd no way that d4, c4 and NC3 deserve negative >assessments. How do you get accurate and correct informant rates or numeric >assessments in there? You need a whole whack of games to do this successfully. You might consider using what is publicly available from the C.A.P. project... and import those epd values into a massive book or something. >Same thing happens when i backsolve a whole database of Caps games, for >instance after e4e5, Nf3 Nc6, d4 gets a negative because the game was lost >by Cap in 1901 when he played Principe in Havana. OK, that was that game, >but there is no way that d4 is a bad move or that it leads to a losing line. >Cap lost the game much later on. So what do you do? Annotate the game with lots of side-variations and assessments, then backsolve again and the backed-up values will be more realistic. Bookup is just going by the data that you give it. If it isn't told that Capablanca had an improvement on move 27 that would have allowed him to draw, Bookup's backsolving can't take that into account. Dave
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.