Author: Mark Young
Date: 14:13:10 03/19/98
Go up one level in this thread
On March 19, 1998 at 14:36:43, Howard Exner wrote:
>On March 19, 1998 at 13:03:11, Sylvain Renard wrote:
>
>>
>>On March 15, 1998 at 22:50:22, Mark Young wrote:
>>
>>>The Match will end when one program wins 10 games draws not counting.
>>>If the score is 9-9 the match will end as draw.
>>I think that the number of games played will be too low to have a
>>scientific
>>result.
>>I have started a match between the Fritz4.01 and Fritz 5 engines,
>>same opening books , PP 180 MHz , 32 MB of hash tables for each
>>program, "blitz" 1 hour.
>> After 40 games, there score was 20 - 20.
>> But after 94 games Fritz 4.01: 42,5 Fritz 5: 51,5 (20 - 29 - 43
>>draws).
>
>How many double games occured in this 94 game match?
>
>>Conclusion: when the ratings of the programs are close, you have to
>>play many many games before you can say that one is SLIGTHLY better
>>that the other. As I think that ALL programs ahead in the Sweedish
>>Rating
>>List are close, THOUSANDS of games are necessary and any conclusion
>>based on 1 game (I have read some of them here...), 11 games or even
>>100 games is RIDICULOUS.
>
>I agree that a large sample size is needed for comp vs comp games. If
>you
>played another 94 games the match could very well be tied again.
I strongly reject the logic that match results and game results are
meaningless.
If this was the case you can only go by your feelings what program is
better. I have been testing computer programs for years and found over
the years that the cream does rise to the top.
Mark Young
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.