Author: Bill McGaugh
Date: 13:49:42 02/21/02
Go up one level in this thread
On February 21, 2002 at 16:36:04, Dann Corbit wrote: >On February 21, 2002 at 16:32:04, Bill McGaugh wrote: > >>On February 21, 2002 at 15:28:13, Thomas Lagershausen wrote: >> >>>Once again. To play matches with 5min/game +3sec is a joke. >>> >>>Chess needs time to find a plan. >> >>Your comment inspired me to do a quick statistical test of the difference >>between the performance of chess programs at lightning speed vs. the performance >>at longer time controls. >> >>I used the two lists on chessfun's site... >> >>To spare you the details, there is not a statistically significant difference >>(at alpha=.05) between the two groups based on the evidence on that site. > >And yet, I have clearly seen engines perform very differently at different >speeds. > >Resp and Gullydeckel overperform at blitz. > >Amy and Beowulf underperform at blitz. > >The chief thing about a bunch of games is: > >You can only draw conclusions about games under the *exact* same conditions. > >You can only draw conclusions within the statistical validity of the available >data (e.g. it is silly to say that x is better than y if x has 2 more ELO points >and the error bar is 500 points). > >IMO-YMMV. All I stated was that, based on the evidence on that site, that is the result...a large collection of games by most of the better programs... My attempts at chessmaster personalites (using selective search=12, mostly), do show increasing strength with longer and longer time controls...based on thousands of games.
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.