Author: KarinsDad
Date: 17:39:16 01/15/99
Go up one level in this thread
On January 15, 1999 at 20:09:45, John Coffey wrote: >Ah the good old days when you could play a chess computer and >move from level 1 up to level 2 etc. Today most programs don't >have levels per se. In fact most just let you set the time control. >This usually means that the program will stomp everybody under >2000 even when you set it to 1 second per move. > >Frizt tries to solve this problem by setting the rating. The lowest >rating seems to be about 1450, and when I tried to play a tournament >game against it I got stomped. Others have had the same experience. > >John Coffey I agree. I have played on "opponents" with ratings slightly higher than my own on CM6000 and the worse case scenario was that I was up two minor pieces and a pawn walking into the middlegame and lost. I consistently can be winning, usually up material, walking into the endgame on an "opponent" 100 to 200 points higher than my rating and get crushed in a matter of minutes in the endgame (this could be due to poor endgame technique on my part, but it doesn't seem to happen in real tournaments). It doesn't matter what amount of time I set it to (usually in the G10 to G30 range), there doesn't seem to be enough time to pull out a win, even in a winning position. I think what happens is that the program randomly picks anywhere from it's best to it's 8th best move. Unfortunately, in an endgame, the 8th best move still seems fairly good. I'm not positive how it is done in CM6000, but if they base it off of the xth best move instead of using a percentage of good moves and a percentage of bad moves in a bell curve based on the rating (i.e. if the best 8 moves are within a half pawn, occasionally it should pick the 9th best move since the best 8 moves are practically the same), then it doesn't work as a playing partner for me. Fritz4 (haven't tried any of the Fritz5s) consistently beats me at a rating 100 points lower than mine. I was planning on putting in the most sophisticated heuristics I could find for determining "average mvoes for a given rating" into my program, just so that I could have a program that I could use as a training partner. I have noticed when examining my games with a chess program that a goodly number of my human opponent's and my moves are often identical (or within the top 3) to what the program would pick, often regardless of playing strength. However, my human opponents and I will "blunder" into lost tactical positions that a chess program will pick up, but we do not pick up over the board. I guess that the programs are just too relentless, even at lower settings, whereas humans have a tendency to swing from great moves to lousy moves with no rhyme or reason to it. The commercial products (and I haven't tried that many, so there could be a good one out there) haven't done it for me. Has anyone else experienced this? KarinsDad
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.