Author: Tony Hedlund
Date: 01:17:00 02/11/01
Go up one level in this thread
On February 10, 2001 at 15:22:56, James T. Walker wrote: >On February 10, 2001 at 12:26:19, Uri Blass wrote: > >>On February 10, 2001 at 11:55:38, Christophe Theron wrote: >> >>>On February 10, 2001 at 08:00:25, Jorge Pichard wrote: >>> >>>>On February 10, 2001 at 01:41:29, Christophe Theron wrote: >>>> >>>>>On February 09, 2001 at 17:11:23, Jorge Pichard wrote: >>>>> >>>>>>I Just checked the SSDF result of the match between Junior 6.0 vs Nimzo 8 on a >>>>>>K6-2 450 MHz and Nimzo 8 was beaten bad by Junior 6.0, by a score of 26.0 to >>>>>>14.0 points including draws. Now this score is compared to my previous match of >>>>>>75 games at G\60 in which Junior 6.0 got the edge by 6.5 points using an AMD >>>>>>K6-2 500 MHz with 128 MB SDRAM. But when I used the Athlon 800 MHz the reverse >>>>>>happened as far as score, not only it happened with my Athlon 800 MHz >>>>>>but in a short match of 20 games at the Cadaque's Tourney, Nimzo also beat >>>>>>Junior 6.0 by 11.5 to 8.5 using a pentium 933 MHz. Therefore, there is a >>>>>>correlation in my judgement in assuming that as the speed of the processor >>>>>>increases Nimzo 8 plays better chess than Junior 6.0. Now I just finished >>>>>>playing 4 games at 40\120 between Deep Fritz vs Nimzo 8 using my Athlon 800 MHz >>>>>>and Nimzo 8 did not benefit from using the Athlon 800 MHz, the score is so far >>>>>>W2 L1 D1 in favor of Deep Fritz. In this case there is also a correlation >>>>>>between the score of Deep Fritz Vs Nimzo as there was on the Cadaque's match to >>>>>>assume that Deep Fritz benefit more than Nimzo 8 as the speed of the processors >>>>>>increases. Lets wait patiently to see what happen as the result is being >>>>>>provided to us daily by Mr.Hans in the SSDF match of Nimzo 8 vs Deep Fritz. My >>>>>>prediction for this match is that it will end up in favor of DF, but by a very >>>>>>small margin. >>>>>> >>>>>>Pichard >>>>> >>>>> >>>>> >>>>>Did I already mention the margin of errors of such short matches? >>>>> >>>>>Oh yes, I did. >>>>> >>>>>And you simply keep on ignoring basic statistics, and you keep on bringing over >>>>>and over your point about Nimzo vs Junior. >>>>> >>>>>Oops... Maybe you have understood basic statistics in a special way? Maybe you >>>>>think that by saying a statement over and over again you can decrease the margin >>>>>of error of this statement until it eventually becomes true? A revolutionnary >>>>>new theory? :) >>>>> >>>>>If you want to demonstrate your point, just play A LOT of games and post the >>>>>result. >>>>> >>>>> Christophe >>>> >>>>75 games are by no mean considered short matches, and your basic statistics are >>>>not precise measurements which equal to chess result. >>>> >>>>Pichard >>> >>> >>> >>>The longer the match, the smaller the error margin. So there is no absolute >>>definition of "short match". >>> >>>What you need to check is the result of the match against the error margin of a >>>match of this length. Then you can tell if the match was "statistically >>>significant", with a given level of confidence, or not. >>> >>>In order to get for example a 90% level of confidence (which still leave 10% >>>chances that what you say is wrong), you'll be surprised by how many games you >>>need to play. >> >> >>I think that all the statistics also may be wrong because the events are not >>independent and the programs use learning. >> >>It is posssible that after 17000 games Nimzo always win Junior because of >>learning or the opposite(I do not know because nobody tried to play more than >>17000 games). >> >>I think that result between 2 programs is not enough to decide which is stronger >>because it is possible that A wins B,B wins C and C wins A. >> >>Uri > >Hmmm. Well 17000 games is an interesting number. I think after that many games >both programs would be reduced to 1 or 2 opening lines and probably playing >draws every time. ;-) Also your statement about 2 programs is probably correct >too. That is why I believe the SSDF should go to a set number of games in round >robin style so that any program is tested the same as all other programs. That >is, they each play the same number of games against the same opposition. >Otherwise you may be skewing the rating by letting one program play more games >vs another program which it has an easier time winning and less games against a >program which it has a hard time to beat. >Jim I thought we were doing that. ;-) Tony
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.