Author: Eelco de Groot
Date: 16:11:38 05/30/01
Go up one level in this thread
On May 21, 2001 at 13:06:51, Andreas Schwartmann wrote:
>I received an e-mail today asking me for help on GS 2930 test suite position
>#13. My chess abilities are not good enough, I am afraid, but what do you think
>about this? Here comes the mail:
>
>Andreas,
>
>I have been using the GS 2930 test suite to estimate the rating for the
>Deep Blue system. This is for use in a paper we are preparing. I
>downloaded the test suite from your web site. There are three problems
>that the system does not solve in 20 minutes. One of them in particular,
>#13, is not solved even for very deep searches. The given move is Bxg7,
>but Deep Blue Jr. gives a different move that has a somewhat higher score
>than Bxg7. The PV follows. Do you have any opinion on this? Thank you.
>
>Regards,
>
>Murray
>
>Rh5 Qc2 Qe7 Qb2 Ne5 Rc1 g4 Rf1 Kf1 Qb1 Kg2 Qe4 Kh3 Qb1 Rf5 Qf1 Kh4 Qg2 h3
>Qa8 Rf7 +1.49
>
>--
>
>Any comments?
>Andreas
[D]2r2rk1/p1q1bpp1/1p6/n2R4/8/P4N2/1B2QPPP/5RK1 w - - 0 1
I let the computer calculate on this position too for a while. Not yet 80 hours
but of course Q5T does not need 80 hours in positions like this.. Seriously
though, as reported here http://www.rebel.nl/ubb/Forum5/HTML/000012.html the
aggressive personalities Q3 an Q5T played Bxg7 in under three minutes on a fast
system and scored over 2700. Still these "personalities" would be rolled up like
a wet newspaper as they used to say over here in Holland, in any serious game.
These are the 11 and 12 ply lines I got to then within 27 minutes on a Celeron
500Mhz:
03:24 11.00 2.73 1.Rh5 Qc2 2.Qxe7 Qxb2 3.Ne5 Rc1 4.g3 Rxf1+ 5.Kxf1 Qb1+
06:36 11.10 2.73 1.Bxg7
15:51 11.10 3.96 1.Bxg7 Kxg7 2.Ne5 Rfd8 3.Qh5 Bf6 4.Nd7 Rxd7
26:58 12.00 3.86 1.Bxg7 Kxg7 2.Ne5 Rfd8 3.Qh5 Bf6 4.Nd7 Rxd7 5.Qg4+ Kf8
I did not repeat this short twenty minute GS2930 calculation but entered 1.Bxg7
and then I got another response, with the same Q5T in Rebel Century 2; it offers
to swap queens first with Black, 1. ..Qc4 instead of taking the Bishop. I have
no idea how bad this move is but I thought it an interesting sideline. The score
for Black with this response is a little better but still at minus 2 or worse. I
feel, it could be a matter of taste here what programs play and it might depend
heavily on what knowledge and bonuses they get for the ensuing endgames. If it
is an endgame the program does not play well, a well-tuned prog would not give
it a high bonus, even if for a GM the ending would give very good chances?
Anyway, after 1.Bxg7 after 22 hours 40 minutes this result hot from the screen
from the Q5T:
Rebel Engine for ECTool. (c) Ed Schröder
Engine version : REBEL CENTURY 2
Rebel personality : "Q5T Personality, very aggressive"
Opening book file : RebelC.mvs
Hash table size : 12 MB
Analysis mode : Analyzing next move
Refresh interval : 500 ms
Last Move : 1.Bxg7 (Black to play)
00:00 03.00 0.71 1..Kxg7 2.Qd3 Qc3 3.Qf5 Bxa3
00:00 05.00 0.63 1..Kxg7 2.Qe4 Qc4 3.Rd4 Qc5 4.Qg4+ Kh8
00:02 06.00 0.12 1..Kxg7 2.Qe4 Rh8 3.Qg4+ Kf8 4.Rd7+ Qc4
00:05 07.00 -0.54 1..Kxg7 2.Qe4 Rg8 3.Ne5 Kf8 4.Qh7 Rg7 5.Qh8+ Rg8
00:14 08.00 -0.95 1..Kxg7 2.Qe4 Rh8 3.Ne5 Nc4 4.Qg4+ Kf8 5.Rc1
00:34 09.00 -1.86 1..Kxg7 2.Ne5 Qc2 3.Qg4+ Kh8 4.Rd3 Qxd3 5.Nxd3 Bxa3 6.Qh5+
03:09 10.00 -2.79 1..Kxg7 2.Ne5 Qc2 3.Qg4+ Kh8 4.Rd3+ Qxd3 5.Nxd3 Rfd8
17:52 11.00 -2.99 1..Kxg7 2.Ne5 Qc2 3.Qg4+ Kh7 4.Rd3 Qxd3
22:07 11.01 -2.99 1..Qc4
29:59 11.01 -2.13 1..Qc4 2.Rd3 Kxg7 3.Nd4 Rc5 4.Qxe7 Qxd3 5.Ne6+ Kh7 6.Nxf8+
51:58 12.00 -2.47 1..Qc4 2.Rd3 Kxg7 3.Ne5 Qf4 4.Rg3+ Bg5 5.Qh5 Qxe5 6.Rxg5+
45:56 13.00 -2.14 1..Qc4 2.Qxc4 Nxc4 3.Bxf8 Kxf8 4.a4 Nb2 5.a5 Rc5 6.Rxc5
bxc5 7.Re1 Bf6 8.Re2
23:07 14.00 -2.67 1..Qc4 2.Rd3 Kxg7 3.Ne5 Qh4
40:36 15.00 -3.01 1..Qc4 2.Rd3 Kxg7 3.Nd4 Bg5 (22:40:36)
---
More interesting perhaps is that Gambit Tiger 2.0 does not want to play either
Bxg7 or Rh5 but eventually 1.Nd2:
00:00:00.7 1,18 6 55860 Rh5 Qc2 Qxe7 Nc6 Qg5 Qxb2
00:00:00.7 1,18 6 57607 Rh5 Rfe8 Re1 Bxa3 Qxe8+ Rxe8 Rxe8+ Bf8 Ne5
00:00:00.2 1,18 7 119363 Rh5 Rfe8 Re1 Qc2 Qe5 Bf8 Qd4
00:00:02.0 1,20 8 359639 Rh5 Rfe8 Nd4 Bf6 Qg4 Nc4 Nb5 Qd8
00:00:06.2 1,12 9 818997 Rh5 Rfe8 Nd4 Bf6 Qg4 Nc4 Qh3 g6 Nb5
00:00:15.8 1,18 9 2125639 Nd2 Bd6 Qg4 Bxh2+ Kh1 f5 Rxf5 Rxf5 Qxf5 Rf8 Qh5 Qf4
Qxh2 Qxd2
00:00:22.4 0,94 10 3124319 Nd2 Bd6 Qg4 Bxh2+ Kh1 f5 Rxf5 Rxf5 Qxf5 Bf4 Qe6+ Kf8
a4 Qc5
00:00:27.1 1,12 10 3737390 Rh5 Rfe8 Nd4 Bf6 Qg4 Nc4 Qh3 g6 Nb5
00:01:10.3 1,18 11 10012596 Rh5 Rfe8 Ne5 Qc2 Qg4 Bf6 Qd7 Bxe5 Bxe5 Qg6 Rh4
00:03:28.5 1,36 12 30603720 Rh5 Rfe8 Bxg7 Kxg7 Ne5 Rg8 Qg4+ Kf8 Nd7+ Ke8 Qxg8+
Kxd7 Rd1+ Kc6 Qxf7
00:16:11.2 1,44 13 140050772 Rh5 Qc2 Qxe7 Qxb2 Qxa7 Qf6 Qd7 g6 Qg4 Nc4 Rc1 Rfd8
Rc2
00:33:13.0 1,68 13 229058433 Nd2 Qb7 Qg4 f6 Re1 Rf7 Rh5 Rc5 Qh3 Rxh5 Qxh5 Nc6
Qg6
00:45:29.8 1,68 14 332367965 Nd2 Qb7 Qg4 f6
01:28:07.7 1,70 15 734997220 Nd2 Qb7 Qg4 f6 Re1 Rcd8 Rh5 Qd7 Qh4 Kf7 Ne4 Rde8
Rh7 Qb5 Bc3 Bxa3
05:40:17.3 1,30 16 -1134496735 Nd2 Rcd8 Qg4 f6 Qe6+ Rf7 Rc1 Qb7 Rxd8+ Bxd8 Nf3
Qe7 Qf5 Bc7 Qg6
After about 24 hours still choosing 1.Nd2 in the 16 ply calculations.
The last line at 13 ply with Rh5 is similar at least in score, 1.44 vs 1.49 to
the DB Jr line: Rh5 Qc2 Qe7 Qb2 Ne5 Rc1 g4 Rf1 Kf1 Qb1 Kg2 Qe4 Kh3 Qb1 Rf5 Qf1
Kh4 Qg2 h3 Qa8 Rf7 +1.49
All in all there is now:
1.Nd2
1.Rh5
1.Bxg7 Kxg7
1.Bxg7 Qc4
Which one of these is best for a computer?
After 1.Bxg7 Gambit Tiger 2.0 plays:
00:00:00.1 0,70 5 5807 Kxg7
00:00:01.1 0,34 6 11598 Kxg7 Ne5 Bd6 Qg4+ Kh7 Qf5+ Kg8
00:00:02.1 -0,15 7 27947 Kxg7 Ne5 Bd6 Qg4+ Kh7 Qf5+ Kg8 Qg5+ Kh7 Qf5+
00:00:06.1 -0,15 8 70091 Kxg7 Ne5 Bd6 Qg4+ Kh7 Qf5+ Kg8 Qg5+ Kh7 Qf5+
00:00:23.6 -0,22 9 280570 Kxg7 Ne5 Rcd8 Qg4+ Kh8 Nd7 f5 Qd4+ Kg8 Nxf8 Rxf8
00:01:02.6 -1,12 10 828549 Kxg7
00:02:30.7 -1,38 11 2081592 Kxg7 Ne5 Rcd8 Qg4+ Kh8 Nd7 f5 Qh5+ Kg7 Nxf8 Rxf8
Rxf5 Rxf5 Qxf5 Bxa3
00:40:31.0 -1,02 12 23970526 Kxg7 Ne5 Rfd8 Qg4+ Kf8 Nd7+ Rxd7 Rxd7 Qc4 Qf3 Nc6
Qh5 Qh4 Qf5 Qg5
02:15:57.6 -1,10 13 88179102 Kxg7 Ne5 Rfd8 Qh5 Bf6 Qg4+ Kf8 Nd7+ Rxd7 Rxd7 Qe5
Rxa7 Rc2 Ra8+ Ke7 Qb4+ Qc5
06:22:04.1 -1,34 14 311209355 Kxg7 Ne5 Rfd8 Qh5 Bf6 Qg4+ Kf8 Nd7+ Rxd7 Rxd7 Qe5
Rxa7 Rd8 Qb4+ Kg7 Qxb6 Nc4
If the score would also be about +1.34 minus something for asymmetry for White
then there is only a small difference with Nd2 that gets 1.30 and 1.Rh5 that
gets a little less. From Gambit Tiger's viewpoint that is.
Please don't look at the NPS of Tiger after 1.Bxg7, as Rebel was running first
it only got to 12000 NPS. As I got only very low NPS maybe somebody could repeat
this, with more hashtables than 13 MB too..
Regards, Eelco
I'm curious if the DB team will read all this.. Their 11 out of 14 is impressive
of course.
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.