Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: a little statistics - sometimes I can't resist :-)

Author: Robert Hyatt

Date: 17:03:02 10/22/02

Go up one level in this thread


On October 22, 2002 at 15:55:44, Christophe Theron wrote:

>On October 22, 2002 at 05:57:23, Brian Katz wrote:
>
>>On October 22, 2002 at 04:52:07, Uri Blass wrote:
>>
>>>On October 22, 2002 at 03:30:12, Stefan Zipproth wrote:
>>>
>>>>On October 21, 2002 at 07:53:53, Brian Katz wrote:
>>>>
>>>>>On October 21, 2002 at 07:40:54, Kurt Utzinger wrote:
>>>>>
>>>>>>It is well known that Deep Fritz 7 needs fast hardware to play at full strength.
>>>>>>And in this respect your P_II/350 MHz is indeed too slow and any comparison with
>>>>>>Fritz7 only a waste of time.
>>>>>>Kurt
>>>>>
>>>>>Thank you Kurt
>>>>>That is the answer I was looking for. I suspected that that might be the case.
>>>>>Thank you and Uri for your replies.
>>>>>Much appreciated.
>>
>>
>>>>>Brian
>>>>
>>>>... no, that's not the answer. 8 games say nothing. Like it was said before, try
>>>>tossing a coin 8 times. Both sides have the same "winning" chances, but you will
>>>>easiliy get results like 5-3. To measure the difference between these two
>>>>engines significantly, you would need to play thousands (!) of games,
>>>>independend from time controls. This is simple math, but unfortunately no one
>>>>seems to believe it.
>>>>
>>>>Just use ELOstat - or play another 8 :-)
>>>>
>>>>Stefan
>>>
>>>8 games do not prove which program is better but they may suggest some
>>>conjectures.
>>>
>>>It is a waste of time to play some thousands of games instead of checking the
>>>number of nodes of Fritz at slow hardware and fast hardware to find out if Deep
>>>Fritz7 does not earn more from fast hardware.
>>>
>>>I have not both programs so I cannot do the comparison on fast hardware.
>>>The poster gave some information about the number of nodes in his slow hardware.
>>>
>>>Uri
>>
>>Hello again URI
>>
>>Please tell me why all these other posters think that I am only basing my
>>findings on 8 games. I have run many many tournaments with all different time
>>controls and hash settings. A good number of short tournaments,( approx. 20)and
>>quite a few long 20 game matches. I have also had quite a few 20 round
>>tournaments with many chess engines, not only DF7 vs F7   I have had Deep-Fritz
>>which is (based on Fritz6) come out ahead of DF7.
>>The simple fact of the matter is that DF7 does not perform as well on my
>>computer as Fritz 7.
>
>
>
>My deepest feelings here go to Frans.
>
>I feel exactly the same when I read this as when I read that "Chess Tiger 15
>does not perform as well as Chess Tiger 14 on my computer".
>
>DF7 is definitely stronger than F7, and CT15 is definitely stronger than CT14.
>
>But well... If you do not FEEL that it is the case, what can we do?
>
>
>
>    Christophe
>
>
>
>
>
>
>>This is not based on just engine tournaments, this is also based on the play on
>>my computer account on ICC. I find DF7 struggling through every game. This is no
>>longer a coincidence.
>>I know that 8 games as well as anybody do not mean anything. I have had engines
>>matches where an engine wins the first 4 games in a row in a ten game match only
>>to lose the match in the end.
>>I know  I believe it was Bob Hyatt who said that you need at least 700 games
>>between engines to get an accurate picture.
>>
>>I am just reporting what I have found thus far.
>>I agree that this must be a hardware problem.
>>I don't think I need 700 games when DF7 is losing almost every tournament I have
>>run. It had won a few.
>>DF7 on my computer, clearly does not reflect what Chessbase boasts in relation
>>to DF7's increased positional knowledge and endgame knowledge and increased
>>playing strength over Fritz 7 when used on only a single processor.
>>
>>So please, who ever replies to this post. MY findings are not based on only 8
>>games, they are based on many. At least 200-300 games.
>>What I found odd in the last tournament I ran is the extremely high hash table
>>settings for Fritz 7. I thought that this would handicap Fritz 7 but it still
>>performed better anyway.
>>
>>Brian Katz



It is also possible that Deep Fritz 7 is better on multiple-cpu machines only.
IE the
improvements might have slowed it down a bit on one cpu, but made the parallel
search
more efficient.  If it isn't being used on a multiprocessor, there is little
reason to own a
"deep" program, generally...



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.