Author: Albert Silver
Date: 08:13:56 09/09/04
Go up one level in this thread
On September 09, 2004 at 09:58:09, Stuart Cracraft wrote:
>It seems to me that the bulk of this board's contents these
>days is non-programmer, i.e. "I have this program which I didn't
>write and I am going to play it against that program which I
>didn't write so I can prove absolutely nothing but look like a
>computer chess programmer in the process, which I'm not."
Not being a programmer, I know which group I'm in... :-)
I disagree with your conclusions regarding what people post and why. There are a
wide variety of computer-chess topic related posts here, and they are certinaly
ot divided (IMHO) ito programmers and people who 'want' to be a CC programmer. I
know I don't.
For example, let's take the tinkering of settings. Perhaps the closest one could
get to that 'wannabe' argument of yours. I tinker plenty with Pro Deo's settings
too mind you, so maybe that would make you think I do. Hardly though. I am a
chess player first and above all as far as computer chess is concerned. I love
the variety of engines, the variety of feedback, and just watching two strong
and interesting players duke it out. Playing with the settings allows me to
change its behaviour, produce very different gamnes, and who know? Possibly even
better results overall. It would hardly be the first time. A number of better
performing settings have been found for CM9000, a better group was found for
Hiarcs 8 (called the 'Bareev' style), so if this helps great. Either way, its
fun, that's all.
How about the engine matches? Well, I have no doubt you, and all authors do
plenty of testing of their own, yet unless you are very wealthy or have access
to a LOT of computer time and power, it is probably not easy to get enough good
information. So, I'm sure that this may help other authors. Even if it doesn't
actually help them, if I post results of a match, testing conditions or engine
strength, I really fail to see how this would make me look like a programmer or
even be an attempt to do so. The holy grail of who is strongest and by how much
is of constant interest to many. The SSDF, even when they were running,
essentially tested the commercial programs. It's plenty of work, so I hardly
blame them for not including 50 other amateurs, but the question on those
amateurs remains unanswered all the same. I don't see anyone ever posting
matches between Junior 8 and Deep Fritz 7 for example, reproducing the results
of the SSDF, so new and more info is sought after. How they do in blitz or
bullet conditions, and how they fare against the amateurs.
What about the positions then? Who cares if your engine can or can't find the
solution to position XYZ, and if some souped up Shredder can in 0.001 seconds?
Well, for the non-programmers, it represents two things. At least that is how I
see it. It is an interesting position to solve. As a human. I have taken more
than one position posted here, and shared it at my club in one of my lectures on
tactics. As to computer programs, it helps highlight weaknesses and strenghts.
Knowing an engine's Elo, no matter how precise it may be, says little about what
it does well or not. As to the programmers, it may help highlight issues they
themselves may wish to address. For example, take a look at this endgame
position I ran into in a blitz game between Ruffian (disproportionately stronger
at blitz BTW) and some Pro Deo settings I was testing:
[D]8/7b/7P/3p4/1Kp1k3/P1B5/8/8 b - - 0 71
As far as I can tell, there is only one drawing move here: 1...d4! allowing the
bishop to come and stop the a-pawn. 2.Kxc4 dxc3 or 2.Be1 c3 3.a4 Kd5 Other moves
I analyzed lose:
a) 1...Kf5? to bring the king to the h-pawn and allow the bishop to cover the
a-pawn. No winning attempt of course, but it doesn't even draw because white
will protect the h-pawn with the bishop, and win the black bishop with the
a-pawn. As the remaining bishop is the correct one for the h-pawn, white wins.
Ex: 2.a4 Bg8 3.a5 Kg6 4.a6 d4 (forced) 5.Bxd4 Bd5 (if 5...Kxh6 6.Kc5) Be3 +-
b) 1...Bg8? hoping to still allow a timely d-pawn sac. 2.Kc5! and the same
problem occurs again as I am controlling the d5 square with the king now.
c) 1...Kd3? This is what Pro Deo played, but as said, unless it's d4 it is just
a choice between losing moves. 2.Bf6 d4 3.a4 c3 4.a5 c2 5.Bg5 and the bishop on
h7 can't stop both a- and h- pawns.
A pure search, no matter how fast, seems insufficient to see this, so you either
have 6-piece tablebases like Bob or you have the knowledge hardwired into the
engine. It's not self-evident at all. Pro Deo obviously didn't have it, but some
of the commercial stumble on it too:
Analysis by Deep Fritz 8:
1...Kd3 2.Bg7 Be4 3.a4
= (0.21) Depth: 8/16 00:00:00 14kN
1...d4!
= (0.20) Depth: 8/16 00:00:00 15kN
1...d4!
= (0.05) Depth: 8/16 00:00:00 16kN
1...d4 2.Kxc4 dxc3 3.Kxc3 Kf5 4.a4 Kg5 5.a5 Kxh6
= (0.00) Depth: 8/16 00:00:00 18kN
Analysis by Junior 8:
1...Kd3 2.Bh8 d4 3.Kc5 c3 4.Bxd4 c2 5.Bb2 Kd2 6.a4 c1B 7.Bxc1+ Kxc1
= (0.00) Depth: 18 00:00:00 341kN
1...Kd3 2.Bh8
² (0.30) Depth: 21 00:00:02 2949kN
1...d4 2.Kxc4 dxc3 3.Kxc3 Kd5 4.Kb4 Kc6 5.Kc4 Bg8+ 6.Kd4 Kb5 7.Kc3 Ka4
= (0.16) Depth: 21 00:00:03 3906kN
1...d4 2.Kxc4 dxc3 3.Kxc3 Kd5 4.Kb4 Kc6 5.Kc4 Kb6 6.Kc3 Kb5 7.Kb3 Bg8+ 8.Kc2
Bh7+ 9.Kc3 Ka4
= (0.07) Depth: 24 00:00:10 13883kN
But.... look at a couple of the endgame experts:
Analysis by Hiarcs 9:
1...Bg8 2.a4 d4 3.Bb2 Kd5 4.a5 d3 5.Kc3 Kc6 6.Kd2 Bh7 7.Bc3 Kd5
= (0.18) Depth: 11/29 00:00:14 4786kN
1...Bg8 2.Kc5 Kd3 3.Bd4 c3 4.a4 c2 5.Bb2 d4 6.a5 Ke3 7.Bc1+ Ke2
= (0.23) Depth: 12/30 00:00:29 9431kN
1...Kd3 2.Bf6 d4 3.a4 c3 4.a5 c2 5.Bg5 Ke4 6.a6 Kf5 7.Bc1 Kf6 8.Kc5 Ke5 9.a7 Be4
= (0.20) Depth: 12/30 00:00:31 10265kN
1...Kd3 2.Bf6 d4 3.a4 c3 4.a5 c2 5.Bg5 Ke4 6.a6 Kf5 7.Bd2 Kf6 8.Kc5 Be4 9.Kxd4
² (0.28) Depth: 13/30 00:00:44 14563kN
1...Bg8 2.a4 d4 3.Bb2 Kd5 4.Ba1 d3 5.Bc3 Kc6 6.a5 Kb7 7.Kb5 Ka7 8.a6 Ka8 9.Kc5
= (0.24) Depth: 13/30 00:01:18 25438kN
1...Bg8 2.Kc5
² (0.49) Depth: 14/31 00:05:34 96758kN
1...Bg8 2.Kc5 Kf5 3.a4 Kg6 4.a5 Be6 5.a6 Bc8 6.a7 Bb7 7.Bd2 Ba8 8.Kd4 Kh7 9.Bf4
² (0.58) Depth: 14/31 00:06:35 117595kN
1...Kd3 2.Bf6 d4 3.a4 c3 4.a5 c2 5.Bg5 Ke4 6.a6 Kf5 7.Bd2 Kf6 8.a7 Be4 9.Kc4 Kg6
10.Kxd4
² (0.43) Depth: 14/31 00:07:12 129531kN
Analysis by Shredder 7.04:
1...Bg8 2.Kc5 Kf3 3.a4 Be6
± (1.13) Depth: 18/37 00:00:18 11813kN
1...Kf5 2.Kc5 Kg6 3.Bd2 Bg8 4.a4 c3 5.a5 cxd2 6.a6
± (1.12) Depth: 18/37 00:00:19 12572kN
1...Kf5 2.Kc5 Ke6 3.a4 Be4 4.a5 Kd7 5.a6 d4 6.Kxd4 Bb7 7.axb7 Ke6 8.b8Q Kf5
9.Kxc4 Kg6
± (1.11) Depth: 18/37 00:00:19 12903kN
1...Kf5 2.a4 Kg6 3.Bd2 Bg8 4.Kc5 c3 5.Bxc3 Kxh6 6.a5 Be6 7.Bg7+
± (1.28) Depth: 19/35 00:00:25 16897kN
1...Kf5 2.a4 Kg6 3.Bd2 Bg8 4.Kc5 c3 5.Bc1 Be6 6.a5 Bc8 7.Kxd5
± (1.27) Depth: 20/38 00:00:37 24708kN
1...Kf5 2.a4 Kg6 3.Bd2 Bg8 4.Kc5 c3 5.Bf4 Kf5 6.Bc1 Kg6 7.a5 Bf7
± (1.36) Depth: 21/41 00:01:02 40450kN
1...Kf5 2.a4 Kg6 3.Bd2 Bg8 4.a5 Be6 5.Kc5 Bc8 6.Kxd5 c3 7.Bf4 Kh7 8.Kc6 c2
± (1.38) Depth: 22/42 00:01:55 73347kN
1...Kf5 2.a4 Kg6 3.Bd2 Bg8 4.a5 Be6 5.Kc5 Bc8 6.Kxd5 c3 7.Bf4 Kh7 8.Kd4 c2 9.Kc3
Bf5 10.a6 Kg8
± (1.38) Depth: 23/44 00:03:10 118104kN
Crafty gets this instantly too BTW with no tablebases. I checked.
New game
8/7b/7P/3p4/1Kp1k3/P1B5/8/8 b - - 0 1
Analysis by Crafty 19.15:
1...d4
= (0.17) Depth: 1/5 00:00:00
1...d4 2.Bd2
= (0.04) Depth: 2/5 00:00:00
1...d4 2.Kxc4 dxc3 3.Kxc3
= (0.00) Depth: 3/5 00:00:00
1...d4 2.Kxc4 dxc3 3.Kxc3
= (0.00) Depth: 4/7 00:00:00
1...d4 2.Kxc4 dxc3 3.Kxc3 Bf5
= (0.00) Depth: 5/7 00:00:00
1...d4 2.Kxc4 dxc3 3.Kxc3 Bf5 4.Kd2
= (0.00) Depth: 6/9 00:00:00 1kN
1...d4 2.Kxc4 dxc3 3.Kxc3 Bf5 4.Kd2 Bh3
= (0.00) Depth: 7/9 00:00:00 2kN
1...d4 2.Kxc4 dxc3 3.Kxc3 Bf5 4.Kd2 Bg6 5.h7 Bxh7
= (0.00) Depth: 8/11 00:00:00 6kN
So the biggest issue here is one of knowledge and cretainly this can be of use
to programmers, no?
Would I want to be a computer engine programmer? Frankly, no. I am well aware
the time this requires and work, and while I admire your efforst and will gladly
cheer you on, I have other interests that attract me far more.
Albert
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.