Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF Rating List 2006-02-10

Author: Uri Blass

Date: 11:34:02 02/11/06

Go up one level in this thread


On February 11, 2006 at 14:20:59, Uri Blass wrote:

>On February 11, 2006 at 13:40:02, Chessfun wrote:
>
>>On February 11, 2006 at 04:33:59, Uri Blass wrote:
>>
>>>On February 11, 2006 at 04:19:04, Chessfun wrote:
>>>
>>>>On February 11, 2006 at 04:07:35, Chessfun wrote:
>>>>
>>>>>On February 10, 2006 at 18:13:33, Uri Blass wrote:
>>>>>
>>>>>>On February 10, 2006 at 17:30:10, Roger D Davis wrote:
>>>>>>
>>>>>>>On February 10, 2006 at 15:55:55, Russell wrote:
>>>>>>>
>>>>>>>>I dont agree with the list. H10 Hypermodern is not stronger than Fritz 9 nor is
>>>>>>>>S9 sronger than Fritz 9.
>>>>>>>
>>>>>>>
>>>>>>>They're testing with 1200 megahertz AMDs, right?  Perhaps at that low a speed,
>>>>>>>the rankings of the programs breaks out differently.
>>>>>>>
>>>>>>>Roger
>>>>>>
>>>>>>Low speed?
>>>>>>
>>>>>>120/40 for 1200 Mhz AMD with ponder on is clearly higher level than 40/40 for 2
>>>>>>ghz with ponder off(CEGT conditions).
>>>>>>I do not see reason to defend Fritz.
>>>>>>People who complain probably never tested it at slow time control and tested it
>>>>>>mainly at blitz.
>>>>>
>>>>>CCRL tested it at 4040 on AMD 3800+ and it is stronger IMO than shown by the
>>>>>SSDF results. I think it will increase in the SSDF as more games are played.
>>>>>http://kd.lab.nig.ac.jp/chess/CCRL-4040/rating-table-all.shtml
>>>>>
>>>>>>It is possible that the picture in blitz is simply different and blitz on faster
>>>>>>hardware is not equivalent to 120/40 on A1200
>>>>>>
>>>>>>Note that comparing results of CEGT with SSDF also does not make sense because
>>>>>>conditions are different.
>>>>>
>>>>>Then why compare them? The poster simply stated that Fritz is stronger than as
>>>>>shown he never mentioned at blitz.
>>>>>
>>>>>>Different opening book
>>>>>>Different ponder on/ponder off
>>>>>>
>>>>>>This may be enough to explain slightly different results.
>>>>>
>>>>>If you now wish to look at the different results of SSDF and CEGT of course the
>>>>>different books, ponder and large time control difference is alone enough of an
>>>>>explaination for differences. But again the original poster was never comparing
>>>>>either.
>>>>>
>>>>>Sarah
>>>>
>>>>I could also add the latest selective search rating lists indicate that Fritz 9
>>>>is stronger which may have also been something considered by the original
>>>>poster. Whereas you assumed a comparison with blitz controls.
>>>>
>>>>PC PROGRAMS
>>>>Ratings are calibrated to Pentium4 machines running at 1200MHz
>>>>
>>>>.............BCF Elo.. Program
>>>>
>>>>268 2755 Fritz9 (nearly 400 games played)
>>>>267 2742 Fruit2.2 WCCC05 (nearly 600 games played)
>>>>267 2737 Shredder9
>>>>266 2732 Shredder8
>>>>263 2705 Junior9 (Deep Junior9 add +40 Elo on dual/1000, +70 on quad/1000))
>>>>261 2693 Junior8
>>>>260 2688 Fritz8. (Deep Fritz7/8 add +40 Elo on dual/1000, +70 on quad/1000);
>>>>Fritz8-Bilbao probably +20 Elo
>>>>260 2683 Hiarcs9 (Hiarcs9.6 MAC appears to be +30/40 Elo improvement, Hiarcs10
>>>>out soon)
>>>>
>>>>Sarah
>>>
>>>Pentium4 is known to be slower for chess relative to A1200
>>>I also remember from the past that they used games from other sources that were
>>>not in SSDF conditions and I simply do not take the selective search rating
>>>seriously and it seems to me that they simply combine results of different tests
>>>that were not in the same conditions.
>>
>>
>>As you state yourself pentium 4 is slower relative to A1200 but they also use
>>SSDF results.
>>
>>There are many rating lists across the internet that combine others results with
>>their own to produce a greater list. But to say they combine results of
>>different tests seems to be to be wrong. They are specific of stating what they
>>test but the choice as with anything is of the reader.
>>
>>Anyway there conditions are:
>>
>>"Each computer/program's rating figure is calculated by combining its
>>computer-v-computer results WITH its results against graded players in proper
>>tournaments. The level of the finished List is determined by the results v
>>graded players, with a strong bias towards most recent results. Considerably
>>more computer-computer games are played than those v. humans, as it is
>>physically and financially impossible to enter enough Tournaments to obtain the
>>number of games v. humans that we would wish.
>>
>>Gradings from computer v computer results do not always match those of computer
>>v humans: a program CAN be prepared specifically to do its best against other
>>computers, OR it can be programmed to do its best against humans! The latter is
>>what USERS want, but programmers inevitably have an eye on our Rating Lists, and
>>are aware of the preponderance of computer-computer games which produces them!
>>Even so, my view is that it is BETTER to have this type of Computer Rating List,
>>as a guide for interested parties and prospective purchasers, than to rely
>>solely on manufacturers' claims! Another factor is that Chess Players are MUCH
>>more used to playing against Computers and Computer Programs than they were 5,
>>10, 15 years ago. Programs like the Fidelity Sensory9, Novag Super
>>Constellation, Mephisto Academy, Mephisto Lyon 68030 etc. would NOT be able to
>>obtain the same gradings they did in the 'old' days were they entered in the
>>same Tournaments again today!
>>
>>The results on which my Ratings are based include my own testing, that of
>>Selective Search readers, and (with permission) the valuable SSDF scores. So the
>>calculations combine MANY results and I believe that they achieve a high level
>>of accuracy. Time controls used are from a minimum G/60 or 60/60 up to and
>>including full Tournament time controls (as used by the SSDF) 40/2hrs. The
>>initial calculations are done in Elo, and converted to British BCF figures using
>>the formula (Elo-600)/8 = BCF."
>>
>>Sarah
>
>In the specific case I am almost sure that they had no ssdf games at the last
>time that it was updated.
>
>The ssdf used fruit2.2.1 and they used fruit2.2 or FruitWCCC
>
>They probably used mainly CEGT games that were the first available games for
>fruit and they were without original books.
>
>The first game of fruit with original book that I read in this forum had clearly
>better results and I remember from this forum(not from the ssdf results)
>something like 28-14 for fruit2.2 against fritz9 in the time fruit came out.
>
>Uri

Note also that the difference based on the selective search list between fritz
and shredder is only 18 elo.

change in condition:
yes ponder/no ponder may be clearly enough to explain difference of 20 elo even
without statistical error.

My point is that it is only a guess that Fritz9 is better than Shredder9 if
enough games are played and we have no evidence that it is the case for ssdf
conditions.

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.