Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: WOW {Yawn}

Author: Dann Corbit

Date: 15:20:06 07/01/02

Go up one level in this thread


On July 01, 2002 at 17:22:45, stuart taylor wrote:
[snip]
>At 500 elo difference, one win (to the lower rated one) should be quite an
>interesting happening (two draws would be more likely).
>Two consecutive wins should be extremely unusual and a little suspect.
>Three consecutive wins should be plenty cause for shock. Especially due to the
>fact of them being consecutive, and the only games so far.
>Four consecutive wins should be plenty grounds to consider it proven to be well
>under 500 elo difference, or, some other equipment failing or virus.

That's absurd.  Look at the error bars and you will see your term "proof" go
poof.

>I would never say such a thing if it were 4 wins amongst 7-8 games, although it
>would normally be spread out amongst about 80 games. I think. But consecutivity
>is very meaningful mathematicaly.
>Machines don't have moods.

No but they do have bugs.  And there may be a bad line exploited over and over
again (intentionally or otherwise).  I have often seen the case where one engine
with a superior ELO gets clobbered by a vastly inferior engine.

>The wins being in blocks should in general look more
>suspect than they are. Unless they are closely rated.

I am sure you are right, but how big of a block is suspect?

I am running a big contest now, with two game matches between each pair of
engines.  Eventually, we will have 4 games for each pair.  Have a look at this:

[Event "Computer chess game"]
[Site "MWMCKEE"]
[Date "2001.10.01"]
[Round "1"]
[White "ZChess-120"]
[Black "LarsenVB-407"]
[Result "0-1"]
[ECO "C00v"]
[Variation "French: 2.d4"]
[TimeControl "7200"]

1.d4 e6 2.e4 Bb4+ 3.c3 Be7 4.Qg4 g6 5.Bb5 Nf6 6.Qe2 a6 7.Bd3 Nc6 8.Bh6 d6
9.h3 Bf8 10.Bxf8 Kxf8 11.Nd2 d5 12.e5 Nh5 13.Qe3 f6 14.g4 Ng7 15.Ngf3 fxe5
16.dxe5 Kg8 17.h4 Bd7 18.h5 gxh5 19.gxh5 h6 0-1

[Event "Computer chess game"]
[Site "MWMCKEE"]
[Date "2001.10.01"]
[Round "2"]
[White "LarsenVB-407"]
[Black "ZChess-120"]
[Result "1-0"]
[ECO "A06"]
[Variation "Reti: 1...d5"]
[TimeControl "7200"]

1.Nf3 d5 2.e3 Nf6 3.Bb5+ c6 4.Be2 Bg4 5.d4 e6 6.Nbd2 Qc7 7.h3 Bh5 8.g4 Bg6
9.g5 Ne4 10.h4 Nd7 11.h5 Bf5 12.Nxe4 dxe4 13.Nh4 Be7 14.Nxf5 Qa5+ 15.c3
Qxf5 16.h6 g6 17.Qb3 1-0

Zchess is far stronger than Larsen VB.  In the current rankings...

Rank Program           Elo    +   -   Games   Score   Av.Op.  Draws
...
 17  ZChess-120      : 2450   35  67   197    73.9 %   2270   15.7 %
...
 72  LarsenVB-407    : 2020  333  95    27    11.1 %   2382    0.0 %

If you run enough matches, you will see far stranger things still.



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.