Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: only 179 left after cleaning up all buggy results!

Author: Jonas Cohonas

Date: 05:06:49 01/23/06

Go up one level in this thread


On January 23, 2006 at 06:08:13, Günther Simon wrote:

>On January 23, 2006 at 03:01:16, Jonas Cohonas wrote:
>
>>Rank	Engine		                          Score           %
>>1	Rybka v1.01 Beta 11.w32	 	 	138.5/220	62.9	· ······
>>2	Shredder 9.1UCI 	 	 	41.5/110	37.7	35-62-13
>>3	Fruit2.2.1 	 	 	        40.0/110	36.3	26-56-28
>>
>>I had set it to play 220 games based on the Rybka 220 opening positions (for
>>testing) .epd which means that they made it to position #54.
>>
>>For some odd reason they played 221 games and there where a few games ended due
>>to "illigal move" on the first move and it always seems to happen when the first
>
>It did not happen always in the first move! Sometimes it happened 10 moves or 15
>moves later and sometimes it happened even in move 60 or 70.
>Also those are not a 'few' games. After eliminating 41! buggy illegal move
>results and a few buggy conflicting results, only 179 of 220 games were left.
>And even those still contained at least 2 time losses of Shredder despite
>an increment tc(180+2)?

Yeah that is odd indeed.

>>move made was a knight move, if this is due to a bug in arena (Arena 1.99beta2),
>
>Have you started with beta 1 and later changed to beta2?
>I thought all those results bugs were fixed meanwhile? I wonder
>why all the last 45 or so games had no buggy results?
>Note also that the Arena site says those new betas are only
>for testing and shouldn't be used for games(posted somewhere).

Nope all in the same version.

>>the .epd positions, one of the engines i don't know, but after a quick scan it
>>seems evenly devided, however this makes the result a little bit unreliable, but
>
>A quick scan is not enough, if you want to post reliable results.
>Also, _if_ all the buggy results would have been 'evenly' divided
>(which they aren't), you still had to simply eliminate them.
>Evenly errors only won't change result outcomes between exactly evenly
>programs and this is only the case between identical programs.
>It is very easy and was often told here, because some people e.g.
>defended a very buggy book choice by the same sentence:
>'The buggy lines were evenly divided', that's just nonsense, because
>at least one program (in a match) is handicapped if it is just
>marginally better than the other...
>The same here, buggy results/lines, whatever, cannot be 'evenly' divided.
>Please delete all that buggy results next time, before posting and
>releasing a pgn file.

I wanted to release the whole .pgn as it contained errors ans someone (like you)
could clear up the whole mess.

>After deleting 41 buggy results (still including 2 fishy Shredder
>time losses) the outcome is:
>
>Rybka   118.5/179  66.20%
>Fruit    31.0/ 91  34.06%
>Shredder 29.5/ 88  33.52%

Well that is pretty damn close to the buggy result :)

>>should reflect pretty well the difference in strength.
>>
>>Rybka played with default settings and so did both Fruit and Shredder.
>>
>>For more info and all the games zipped:
>>
>>http://www.fusionweb.dk/chess/rybka/220/
>
>Guenther

Thanks for cleaning the pgn file, any chance you could send me the cleaned up
pgn file?



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.