Author: Uri Blass
Date: 05:20:00 10/14/03
Go up one level in this thread
On October 14, 2003 at 07:58:17, martin fierz wrote: >On October 14, 2003 at 07:00:39, Uri Blass wrote: > >>On October 14, 2003 at 06:46:39, martin fierz wrote: >> >>>On October 14, 2003 at 06:19:17, Gerhard Sonnabend wrote: >>> >>>>Hi ! >>>> >>>>The second match is closed. >>>>(I only post the short tables here) >>>> >>>>Could any chessprogram profit from longer/shorter levels ? >>>> >>>>At the moment i carry out an experiment to find out if there are chessengines >>>>which profit from shorter or longer (time)levels more than other engines. >>>>Played on a P4-1600 / 64MB HTs / 4-TBs / ponder=off >>>>with the "Noomen" (A-H) (=160 games every match) under the ChessBase-Fritz7-GUI. >>>> >>>>Completed are: >>>>Shredder 7.04 vs Junior 8.0.0.2 >>>> Total (+ 1/2 -) >>>> 5min/game 98.5-61.5 (87-23-50) >>>> 10min/game 94.0-66.0 (76-36-48) >>>> 30min/game 92.5-67.5 (70-45-45) >>>>120min/game 89.5-70.5 (65-49-46) >>>> >>>>and: >>>> >>>>Shredder 7.0(CB) vs Fritz 8.0.0.5 >>>> Total (+ 1/2 -) >>>> 5min/game 63.0-97.0 (40-46-74) >>>> 10min/game 68.0-92.0 (49-38-73) >>>> 30min/game 72.5-87.5 (47-51-62) >>>>120min/game 68.5-91.5 (43-51-66) >>>> >>>>The current matches is (after 120 games per Level): >>>>Chess Tiger 15.0(CB) "Normal" vs Beta-WIN-Rebel 12 (style=Test12a) >>>> Total (+ 1/2 -) >>>> 5min/game 83.0-37.0 (68-30-22) >>>> 10min/game 80.0-40.0 (63-34-23) >>>> 30min/game 68.5-51.5 (48-41-31) >>>>120min/game 65.5-54.5 (41-49-30) >>>>(Played on a Cel. 1.8GHz / 128MB HTs / ...the rest look above) >>>> >>>>The details and the games can be found on: >>>>www.pcschach.de >>>> >>>>Best G.S. >>> >>>nice tests! shredder vs junior and CT15 vs rebel 12 beta at least show trends of >>>favoring one program at short and another program at long time control. >> >>It does not prove it. >>It also may show rend of being closer to 50% at longer time control. > >i never said it was a proof :-) >it's a trend, that's all i said. your hypothesis is certainly another one that >waits to be tested! > >>>this seems quite natural to me. if one program has better move ordering than >>>another it will do better at longer time controls, since better MO is an >>>exponential gain. i have seen this behavior in my checkers program in matches >>>against another program - mine does better at longer time controls. and, since i >>>know the other programmer well, i also know that i spend a lot more time on move >>>ordering. so i'm not really surprised about my checkers result. i'm more >>>surprised that some people don't want to believe that this effect exists :-) >>> >>>cheers >>> martin >> >>I agree that if one program is better on move ordering it is going to be better >>at longer time control but the question is practical and not theoretical. >> >>If the better program in move ordering is also better in other things then the >>inferior program is not going to be tested because for some reason only top >>programs are tested and if the top programs earn the same from time we are going >>to find nothing. > >we would find nothing if move ordering (or anything else that influences the >branching factor, like e.g. pruning) were exactly the same for all top programs. >now that is something i would think is *very* unlikely! > >cheers > martin It is not exactly the same but if the difference is small we can find nothing because the statistical error is bigger. Uri
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.