Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: ChessMaster 9000 vs Fritz 7 (10 games PGN)

Author: James T. Walker

Date: 15:12:37 09/17/02

Go up one level in this thread


On September 17, 2002 at 10:39:10, Uri Blass wrote:

>On September 17, 2002 at 09:56:29, pavel wrote:
>
>>On September 17, 2002 at 09:52:49, Uri Blass wrote:
>>
>>
>>>
>>>Game 2 and 6 are the same in the first 35 moves.
>>>
>>>I think that the ssdf should not test chessmaster because it seems to have no
>>>learning function.
>>>
>>>Uri
>>
>>Did CM8k had learning function, does all the programs in SSDF have learning
>>function?
>>If not, than Cm9k should be tested by SSDF.
>>
>>cheers,
>>pavs
>
>The fact that programs in the past were tested without learning does not justify
>continuing this.
>
>The problem is that the result may be dependent more on luck.
>If program A will play 20 games against chessmaster and program B will play
>40 games in the next list then you can expect B to earn rating and A may suffer
>from it.
>
>This does the error in the ssdf list bigger and this is a good reason not to
>test it.
>
>Uri

Hello Uri,
I have said all along this is one of the defects of the SSDF testing.  It does
not apply only to engines without learning.  It also causes errors because some
engines may play better vs one program but not against another.  So if you want
to increase the rating of that program you can just play more games vs an
opponent it plays well against.  The problem with Chessmaster is the decision to
not incorporate learning is up to the programmer.  It is a deficiency which he
has decided to live with.  If Chessmaster gets a lower rating because it's too
stupid to stop playing bad lines then that's too bad.  I guess the programmer
could request that SSDF not test his program if he doesn't want to show it in a
bad light.
Jim



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.