Author: Christophe Theron
Date: 21:53:13 01/31/00
Go up one level in this thread
On January 31, 2000 at 02:59:25, Roger wrote: >On January 30, 2000 at 13:36:38, Hans Christian Lykke wrote: > >> >>The first new program during 2000 is Amir Bans and Shay Bushinskys >>Junior 6.0 K6-2 450 MHz. Junior 6.0 immediately took the first place >>with 2706 after 403 tournament games! This is 77 points more than >>Junior 5.0 has on the same hardware! It's fascinating that it >>still is possible to increase the playing strength of chess >>programs, measured in the way SSDF does. >> >>Compared to Chess Tiger 12.0 Junior 6.0 has 24 more points. Part >>of this difference has been gained due to Chess Tigers tendency >>to repeatedly lose against Junior 6.0 in the same opening. >>Hopefully the opening books and the learners will be improved >>in the future so that these phenomenon disappears. >> >>The order of the programs on K6-2 has changed somewhat on this >>list, thanks to the fact that more games have been played. Chess >>Tiger 12.0 has lost 14 points, Fritz 5.32 12 points, Nimzo 99 >>17 points and Hiarcs 7.32 has 12 points less than on the latest >>list. Junior 5.0 has increased it's rating with 10 points. >> >>The average rating gain between P200 MMX and K6-2 450 MHz is >>now 79 points. This is based on 1537 games with the four programs >>which have been tested on both hardwares. >> >>Next official list will be made in March. >> >>Thoralf Karlsson > >1. Wouldn't Tiger repeatedly lose to ALL the other programs, not just Junior? >There are quite a few games versus Nimzo and Fritz on the 200 megahertz >machnines there, too. Is it observed with them? > >2. Interestingly, the results show that Junior 5 whomps Tiger more than Junior >6, 33 to 9 (j5) versus 24.5-17.5 (J6). What is it about J5 versus Tiger that >creates this finding? Obviously it's not tuning, since J5 predates Tiger by a >lot. > >3. I assume the column just to the right of the rating gives the width of a 95% >confidence interval... If so, then Junior and Tiger are in a statistical dead >heat. I don't think so. If you take a 95% confidence interval, you cannot say which is best. But if you take a 80% or 70% confidence interval the error margins get smaller and you can say which is best. You can say that Junior6 is best whith (approx) 70 to 80% confidence, which is enough for me to say that Junior6 is better than Chess Tiger 12.0 DOS. However I think that Rebel-Tiger, which is faster and has several known improved parameter settings, would be so close to Junior6 that you could really not say which is better. I hope the SSDF will test Rebel-Tiger with its new settings. Christophe >I know it is a linguistic convention to say that the program with the highest >rating is in first place, but conventions aside, mathematically there is no >"first place" here, but only a shared number one: The math is the math (and this >is nothing against Junior. I own Junior 5 and I like its style a lot better than >Fritz 6.0 or Hiarcs 7.32). There is just no #1 here (unless you want to argue >that a smaller confidence interval might be better). > >In this list there are two number 9's, and then an eleven, which strongly >implies that the ratings are more reliable than they really are. > >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.