Author: Roger D Davis
Date: 14:19:50 12/28/05
Go up one level in this thread
On December 28, 2005 at 16:44:21, Joseph Ciarrochi wrote: >Hi folks, I just ran a correlation between the cegt long ratings (40/40 >repeated) and the blitz ratings (40/4). the correlation was extremely high r = >.98, indicating that 96% of the variance in long time control ratings could be >explained by blitz ratings. > >A similar phenomena happens with humans. (i.e., high correlation between blitz >and standard time control ratings). > >Another interesting phenonena is that at higher elo ratings, performance level >does not drop that much between long time controls and blitz (which is why gm's >do so well at simuls). > > > > >Psyshologists have hypothesized that human chess thinking is based on two >things: 1) fast pattern recognition processes (knowledge) and 2) slower search >processes. The very high correlation between blitz and standard time controls, >and the lack of drop in performance at short time controls, have led some >psychologists to conclude that chess skill is largely based on knowledge rather >than on deep search. ( i simplify the theories, but that is about right). If you >reduce the ability to search (as is fairly true in say 3 minute blitz), skilled >chess players (say expert and above) will tend to maintain their relative chess >rankings. > >(of course knowledge and search depth are related, as knowledge should help >increase search) > >Oh, one last interesting finding: As you go down the chess rating ladder, the >nature of the chess thinking changes qualitatively. Lower rated players depend >much more on slow search, and their playing skill is more badly disrupted by >limiting their time. ( in other words, a 2700 rated player would win more games >in a simul against 2300 players, compared to a 1700 player facing 1300 players) > > > >So bringing it back to chess engines.....What distinguishes the playing quality >of one engine from another? Can we apply what we have found with humans (e.g., >it is differences in quality of knowledge)? >Or is computer chess so different as to make any comparison irrelevant? > >best >joseph Have any ideas on the programs for which this prediction fails? For example, what programs are much better at blitz or lightning than would be expected based on their performance at longer time controls? Roger
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.