Author: Chris Carson
Date: 06:00:09 11/19/99
Go up one level in this thread
On November 18, 1999 at 16:43:41, walter irvin wrote: >On November 18, 1999 at 14:35:54, Chris Carson wrote: > >>On November 18, 1999 at 14:06:35, Joshua Lee wrote: >> >>>Hello i was wondering who out there in CCC land is rated the highest would care >>>to do an experiment? If so set Hiarcs 7.32 to a ply depth of 1 and go from >>>there.... play lets say 4 to 10 games at each ply depth and write down your >>>results. This would really tell us all possibly how strong at what ply the comp >>>is. This way the speed of the computer is irrelevant. >> >>I like this idea. I am sure someone will complain about the methodology >>for this experiment (reliability is in doubt), but I think it is a good >>idea. Also, I would recommend that you have a fast machine so that getting >>to higher ply does not take a long time. You could also save time by >>starting at a ply of say 8 and then go up a total of 2 ply and down a total of 2 >>ply (5 ply total with 10 games at each ply, 5 black, 5 white). Any program >>should work. Anyone have crafty for this. If the person is a master, it would >>be great. The chess program is set to ply depth (no time control) and the >>human plays at game in 2 hours (or 40/2 if you want more time). Ponder should >>be off. >> >>Just my two cents. :) My guess: 8 ply = 2000, each additional ply adds >>200 points. I am sure 99% of our readers disagree. :) >> >>Best Regards, >>Chris Carson > >i think you are right or close on the 8 ply , but after that i think the higher >you go you get less per ply . i also believe that a jump from 1 ply to 4 ply is >more than 600 points .because once tactics are handled it comes down to stategy >and the true programs weakness may not be over come so easily just by searching >14 ply instead of 13 .i think the deeper you go each ply may not be worth more >than 50 elo .at some point say after 25 may not be worth more than 5 or 10 >points if that . You raise some very good points. I think some of the computer vs computer (ie crafty vs crafty) that Dr. Hyatt and others have done (see the crafty web site for the results) and published would support your conclusions. I think the research showed that the computer continued to pick a different move with greater depth (ply), but the % that a pv changed was less as the depth increased. I think 15 ply (or about there) has been explored. I agree that deeper search has diminishing returns for computer vs computer, but I wonder if the results might be greater for computer vs human? I do not know of any research on this, especially since GM games are hard to come by. Certainly other factors can provide strength (book, hash, extensions, tb, learning, ...), but it would be interesting to see the effects of depth with the other variables held constant against humans (the stronger the human the better). I think there would be a point where the human gets beat 99.9% of the time and no additional depth will increase the winning percentage (or maybe that would be 100% loss), but I am not sure where the decline occurs and how much the decline is for either humans vs computers or computers vs computers. It would also be interesting to see if a group of humans vs computers produces the same ratings spread vs ply as a single human vs computer. Best Regards, Chris Carson
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.