Author: Frank Schubert
Date: 04:18:38 01/17/00
Go up one level in this thread
On January 16, 2000 at 23:07:12, Len Eisner wrote: >On January 16, 2000 at 21:30:42, James T. Walker wrote: > >>On January 16, 2000 at 18:13:14, Len Eisner wrote: >> >>>On January 16, 2000 at 17:50:33, Frank Schubert wrote: >>> >>>>...after 160 games. I added the result of Shredder - Century 13.5 - 6.5 >>>>(+11,=5,-4). >>>>Here is the new Elo rating list (the old values are given in parentheses). >>>>Again Elo calculation was performed by iteration procedure (start Elo 2600). >>>> >>>> >>>> Program Elo (old) + - Games Score Av.Op. Draws >>>> >>>> 1 Junior 6.0 : 2676 (2665) 101 74 40 60.00 % 2604 45.0 % >>>> 2 Fritz 6a : 2645 (2641) 96 86 40 63.75 % 2546 37.5 % >>>> 3 Hiarcs 7.32 : 2624 (2615) 111 75 40 52.55 % 2606 40.0 % >>>> 4 Shredder 4 : 2610 (2597) 90 66 60 51.67 % 2597 33.3 % >>>> 5 Rebel-Tiger 12.0e : 2603 (2598) 106 59 40 56.25 % 2558 57.5 % >>>> 6 Nimzo 7.32 : 2599 (2592) 76 102 40 41.25 % 2660 42.5 % >>>> 7 Century : 2493 (2493) 73 73 60 32.50 % 2619 38.3 % >>>> >>>> >>>>It is interesting to see that Century's rating did not change. Its Av.Op. is >>>>also still the same. Therefore the result against Shredder fits very well with >>>>Century's results before. Shredder increased its rating by 13 points up to 2610. >>>>Due to the interdependencies of the iteration procedure the ratings of Shredders >>>>opponents are also higher than before. And the ratings of the opponents of >>>>Shredders opponents also increased. Hope I am not confusing you too much but >>>>that's the way it works. You can simply check that the weighted average of all >>>>ratings (weighted by the number of games) is still the same than it was before, >>>>namely 2600. >>>>So no significant changes after this match. Looking for the next one... >>>> >>>>Bye >>>>Frank >>> >>>I calculated the ratings with the Fritz 6 rating function, and the ratings were >>>different. Any idea why? >>> >>>Here are the ratings calculated by Fritz 6: >>> >>>1 Junior 6.0 2406 40 >>>2 Fritz 6a 2391 40 >>>3 Hiarcs 7.32 2372 40 >>>4 Rebel-Tiger 12.0e 2358 40 >>>5 Shredder 4 2351 60 >>>6 Nimzo 7.32 2328 40 >>>7 Century 2242 60 >>> >>>Len >> >>It's simple. Fritz calculates the ELO start list assuming the average ELO is >>2350. This is easy to prove by just taking an average of the results. I'm >>curious as to why 2 programs played 60 games while the others played 40?? >>Jim Walker > >The tournament is still in progress, so some programs have played more games >than others. > >The lower starting ELO would not account for the different position of Tiger and >Shredder on the respective lists. The number of points that separate the >programs is also different. > >Len Hi, I also noticed the differences when I compared my results to those of Fritz. I don't know what kind of Elo formula the ChessBase guys are using. Maybe it's different from mine, but I used the official FIDE formula. But I don't think that's the problem. Concerning the iteration procedure it is easy to check if it's right or not: 1. it must be converging 2. the weighted average of all Elo ratings must be identical with the initial value (I used 2600). (Consider that Shredder and Century played 60 games, the other programs only 40). Both things are fulfilled in my calculations. If I look to the Fritz results I come to the conclusion, that here only the simple average (not the weighted one) is equal to the initial value (2350 here). So who is wrong here ? Let us check both results: The most significant difference is that in my list Shredder is 7 points ahead of Tiger while in the Fritz list Tiger is still 7 points ahead of Shredder. Shredder played against Century (67.5 %), Hiarcs (42.5 %) and Junior (45 %). The mean performance therefore is 51.66667 % which (after FIDE formula) means an Elo difference of +11.59 points against the Av.Op. Tiger played against Century (60 %) and Hiarcs (52.5 %). Mean performance is 56.25 % which means an Elo difference of +43.66 points against Av.Op. Now let's have a look at both rating lists. After my list the average opponent Elo of Shredder is (2493+2624+2676)/3 = 2597.66667. Therefore its final rating should be 2597.66667 + 11.59 = 2609.26 which is equal to Shredders result in my table (2610, differences caused by numerical rounding). Av.Op. Elo of Tiger is (2493+2624)/2 = 2558.5. Therefore its final rating should be 2558.5 + 43.66 = 2602.16 (2603 in my table). So my results are consistent with these considerations. Let's now repeat it with the Fritz table: Av.Op. of Shredder is (2242+2372+2406)/3 = 2340. Therefore its final rating should be 2340 + 11.59 = 2351.59 (2351 in Fritz table, so far calculation is right). But... Av.Op. of Tiger is (2242+2372)/2 = 2307. Therefore its final rating should be 2307 + 43.66 = 2350.66. So it is definitely smaller than Shredders rating and it is different from the result in the Fritz table (2358). So the Fritz result is not consistent with Elo system here. Consequently the iteration procedure in Fritz is maybe wrong. I think their iteration doesn't take into account the different number of games. Could someone be so kind to post the same rating list calculated with ChessBase7 ? Thanks in advance. Bye Frank
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.