Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: tested in various programs

Author: Uri Blass

Date: 14:22:06 08/01/04

Go up one level in this thread


On August 01, 2004 at 16:35:34, Robert Hyatt wrote:

>On July 31, 2004 at 23:14:28, Bruce Humphrey wrote:
>
>>On July 31, 2004 at 16:53:46, Eduard Nemeth wrote:
>>
>>>Question:
>>>
>>>Shredder 8 have 2818 SSDF Elo on AMD 1,2 GHz, right? Yes!
>>>
>>>Which SSDF-ELO have then Shredder 8 on Dual Xeon 3 GHz?
>>>
>>>2900? Or more?
>>>
>>>:-))
>>>
>>>Eduard
>>
>>Yes, a 2900 that is blind to the fact that a rook locked in by pawns is worth 0
>>(or maybe 1) and not 4.88 at depth 24 (I'm sure during the game my Shredder was
>>evaluating 4.6...) as Shredder evaluates in end position in game 1. Well, other
>>software gives even bigger advantage to white!
>>
>>During the game I thought black was going to win the moment Kiriakov plays b5...
>>he did mention just the move before "b5 only plan. White can block with Rb5 and
>>obvious draw", but Shredder is blind to the fact that yes, it wins 1 pawn but in
>>exchange of losing his rooks mobility. The position is really complicated around
>>move 82... what pawn does black take, central or h (via Nf2+-Nxh3)? How does
>>this affect everything? Well, humans (and even more at 6am local time in
>>Krasnoyarsk and with not much time left) tend to have trouble in the
>>complications. A wonderful draw by GM Kiriakov. This makes it (as far as I know)
>>Shredder 8 vs Kiriakov +0-0=4
>>
>>I'm counting Smursky's games too... all 4 games vs dual or better machines.
>>
>>Maybe next one to try should use Tiger or Hiarcs. Or even Gambit Tiger or Junior
>>set in Attack mode. I personally would love it to be Rebel 12. :-)
>>
>>Here is the position tested in various programs... Very curious is the eval
>>difference between hiarcs 9 for PC and Hiarcs 9.46 for Palm!!! I thought they
>>where almost exactly the same, but obviously, 9.46 has been tweaked a lot (in
>>knowledge ?). That, usually, should be for good, but it seems not for this
>>position...
>>
>>comp             depth        eval
>>--------         -----------  ------
>>Fritz 8          depth 20/20  4.38
>>Ruffian ca71     depth 18     4.67
>>Shredder 8       depth 25     4.89
>>Rebel 12         depth 20     4.97
>>wCrafty 19 ca71  depth 18     5.08
>>Gothmog 1 beta7  depth 17     5.40
>>Genius 7         depth 17-28  5.51
>>Hiarcs 9         depth 16/33  5.99 !!
>>Dragon ca71      depth 12     6.02
>>Crafty 19.14 cb  depth 18/25  6.38
>>Junior 8         depth 24     6.39
>>Tiger 15 ca71    depth 20     6.52
>>Tiger 14 cb      depth 17     6.73
>>Hiarcs 9.46      depth 15     7.69 !!   (353sec taken)
>
>
>the data above is flawed in a basic way.  The absolute value of any score is not
>important.  IE I could do a "score = score - 500" in Crafty, and it would play
>_exactly_ the same but with a score that is 5 pawns less than expected.
>
>It is a mistake to compare evaluations.  What is comparable is to pick position
>X, and see what a program thinks, and then position Y and see what it thinks,
>and look at the _difference_ between those to scores as the net gain when going
>from X to Y.  That is what contains actual information, not the static absolute
>score.

It is also not correct because the difference may be bigger if you multiply the
scores by 2 inspite of playing the same moves.

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.