Author: Mike Byrne
Date: 18:24:27 12/23/04
Go up one level in this thread
On December 23, 2004 at 20:08:00, Javier Gamero wrote: >First of all, thanks to Michael Byrne for his work doing "personalities" for >Crafty (and obviously big thank to Bob Hyatt for all his hard, generous work, >past and present). > >I'm interested in Byrne's "special" command "intensity". Reading "readme.txt" >file in Byrne's SE version of Crafty, I can see intensity 1 is about 1200-1300 >ELO and 50 is about 1500-1600. I have some questions about this, mostly for >Michael Byrne, I am afraid :) , but if someone can apport something about it, it >would be very welcome. > >- I assume you are estimating it using a Dual 1.7 GHz, is it right? >- Is Dual 1.7 perfomance nearly the same as a 3.4 Ghz single CPU? No - it would be less and my particular 1.7 Ghz is slower than others - so let's say less than 80% x 3.4 Ghz >- What's Crafty full strength (intensity 10000) in that system? That is just awild guess on my part. I would say that Crafty is ~200 points less than Shredder 7.04 on equal hardware. Then the question is where do you peg Shrdder 7.04 >- Are those ELOs in reference to FIDE ELO or maybe USCF ELO? I am not aware of any valid FIDE or USCF rating for today's programs. Shredder (imo) is equal to at least a top 100 FIDE player and perhaps a top 50 FIDE or even a litte higher on fast hardware. Today a top 50 player is rated near 2650 FIDE. There is no sure fire translation rating table -- but top US players are generally rated higher in USCF by 50 to 150 points. In the USCF, imo, Shredder on fast hardware (say 3.4 Ghz ) would earn over a 2700 USCF rating. When you add significant number of humans to a rating pool, it my belief that the programs rated lower than Shredder would make up some of the gap (between Shredder and Crafty as an example) as computer vs computer has what I call "Bloodgood" effect ratings. Research "Claude Bloodgood" and "+chess +ratings" on google to see how in a small rating pool like SSDF, can distort the ratings differences. I have posted more detailed on this before in CCC and you may find it in the archives. Bottom line, Crafty would be at least 2550 or higher in USCF imo on a 3.4 Ghz and that would place it in a TOP 30 player in the US. > >I know ELO rating of chess "handicapped" engines is just an approximation, but >knowing some data about perfomance a smooth ELO-strength system could be made. >Possibly in weak "personalities" doubling intensity add about 70 ELO and >multiplying by 3 add about 100 ELO (those are known tipical values, although not >an absolute truth, of course). Agreed - I am not sure there is magic mulitplier but it is somewhere between 30 to 100 in many cases and perhaps higher in lower range and lower in the very high end. That is , if we had one single program running on a top end machine- let's say rated 2900 in USCF, and we double the processor speed - I am not sure if we would even get 2925 with the new setup. But if you had a machine rated 2000 in USCF and double the processor speed -I think 2100 is a real possibility. > >I have searched some of the threads about handicapped "personalities" and I find >this question interesting. > >Another question is if Crafty plays "humanlike" chess just with reduced >intensity. I think it is reasonably so, but of course I don't expect it is like >Byrne's weak "personalities". The one un-human like chactertistic when you just play it with less CPU power - it never makes the gross human error - all the errors are really just horizon mistakes - it just did not see far enough. So even at very low CPU power - , Crafty will capitalize on just about every gross human error 9within its horizon). In the real world, humans do not capitalize on every error an opponent makes. > >I think Rebel's play with a chosen ELO is very nice. Also I like Bringer's >reduced ELO feature. Obviously Chessmaster large set of personalities have to be >mentioned, it is good fun for not-so-strong players. Agreed - most of my playing is playing Crafty on a reduced setting. Currently, I play it at 2 seconds per move on my PDA, I beat it occasionaly. It usually sees 5 ply or so and with my rating near 1600 - I would say it is playing near 1900 USCF. Thanks for your questions. Best, Michael
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.