Author: Robert Hyatt
Date: 18:15:00 10/19/00
Go up one level in this thread
On October 19, 2000 at 12:45:51, Gerrit Reubold wrote: >On October 18, 2000 at 14:37:39, Robert Hyatt wrote: > >>Here is my take on computer evaluations: I expect crafty to win all games >>against a GM if the score is (say) 1.5 or higher, and crafty has no more than >>a 1 pawn advantage. That is 1 pawn + a lot of positional edges. I would >>expect to draw positions where the score is 1.5 and crafty has two or three >>pawns. The opponent has a lot of positional compensation. >> >>scores like +2 where it is 2 pawns ahead are hard to measure and could end >>up as wins or draws. > >Hi Bob, > >Do I get you right? > >You (= Crafty) would rather be one pawn up + half a pawn positional score than >to be two pawns up and half a pawn positional compensation for your opponent? > >I am surprised! I thought an "accurate" evaluation should consider both >situations as +1.5, thus exactly equal. Maybe you should consider to scale up >your positional terms (make Crafty more speculative), in order to push Crafty to >positions of the former type!? > >Greetings, >Gerrit Note that what you think I said and what I really said are on opposite extremes of the question of being conservative. I am talking about fractions of a pawn in evaluations, not values in excess of a piece. I base my comment on watching so many games (I would say Zillions but don't want to excite Sarah too much) against GMs and computers and seeing this problem happen over and over. What it really means is that in some cases my positional scores are simply not large enough. IE maybe I shouldn't have gone into a 2 pawn up ending thinking I was 1.5 pawns up. Maybe I should have thought I was only 1.2 pawns up and then I might have gone for a +1.3 ending 1 pawn up. I am generally "comfortable" with Crafty's evaluations. If it is significantly negative, I consider that cause for concern for that side. Ditto when it is significantly positive. It has a pretty good 'feel' for when it is losing and when it is winning, most of the time. Not all of the time, but "most". But taking the scores bigger causes other problems... ie you can have a positional edge that evaporates over time. If you gave up a pawn to get that edge, and it disappears, you are just a pawn down. My Bxa7 trap code is a good example. Most of the time Bxa7 is bad, if the opponent can play b6 and trap the bishop. But not always. And with the big positional edge I give for the bishop trapped, if it gets out, I just end up a pawn down. And probably lose. I'm not ever going to say speculation is bad. I only worry about the magnitude of the speculation. Giving up a pawn is not terribly bad if you get _something_ for it. Giving up a piece is another thing as that is probably going to lose if you are wrong, period.
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.