Author: Eelco de Groot
Date: 16:59:10 12/23/00
Go up one level in this thread
It seems, see below in Jeremiah's post, that Deep Thought-2 needed 6 hours to
come up with Nxh6, searching around 3 million nodes per second.. As Joshua's
Gandalf also showed, these days programs can do better, Q5T needs 1 minute 39
seconds to produce 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 :
Thinking....
+---+---+---+---+---+---+---+---+ Pos 1 - NOLOTP~1.EPD
8 | *R| | | | *Q| *B| | *K|
+---+---+---+---+---+---+---+---+ 00:00:00 00:00:00
7 | | *B| | | | | *p| |
+---+---+---+---+---+---+---+---+
6 | *p| | | *p| *R| | | *p| Time infinite
+---+---+---+---+---+---+---+---+ Hash 40 Mb
5 | | | | *N| | | | | Engine Rebel Century 3.0
+---+---+---+---+---+---+---+---+ Personality Q5T3A.ENG
4 | p | *N| *p| | N | | N | | anti-GM Off
+---+---+---+---+---+---+---+---+ Tactics Off
3 | | | | | | | R | p | Bluff Chess Off
+---+---+---+---+---+---+---+---+ Style Normal
2 | | B | | | | p | p | | Strength Strongest
+---+---+---+---+---+---+---+---+ Search Deeper Search
1 | | B | | Q | R | | K | |
+---+---+---+---+---+---+---+---+
a b c d e f g h
r3qb1k/1b4p1/p2pr2p/3n4/Pnp1N1N1/6RP/1B3PP1/1B1QR1K1 w - -
[Event "?"]
[Site " White to move "]
[Date "2000.12.06"]
[Round "?"]
[White "Pos 1"]
[Black "NOLOTP~1.EPD"]
[Result "*"]
[WhiteElo "?"]
[BlackElo "?"]
[FEN "r3qb1k/1b4p1/p2pr2p/3n4/Pnp1N1N1/6RP/1B3PP1/1B1QR1K1 w - - 0 1"]
1.*
00:00:00 1.00 -10.32 1.Nxd6 Rxe1+ 2.Kh2 Rxd1 3.Nxe8 Rxe8
00:00:00 1.01 -3.62 1.Ne5 dxe5
00:00:00 1.02 -2.88 1.Ngf6 gxf6
00:00:00 1.03 -2.35 1.Nxh6 Rxh6
00:00:00 1.07 -0.76 1.Bc2
00:00:00 2.00 -2.24 1.Bc2 Nxc2 2.Qxc2
00:00:00 2.01 -1.77 1.Ne3 Rxe4 2.Bxe4 Qxe4
00:00:00 2.06 -1.50 1.Nh2 Rc8
00:00:01 2.33 -1.45 1.Bc1 Rb8
00:00:02 3.00 -1.39 1.Bc1 Rc8 2.Nh2
00:00:02 3.01 -1.28 1.Nh2 Rc8 2.Qg4
00:00:02 3.02 -1.28 1.Ne3 Rxe4 2.Bxe4 Qxe4 3.Nxd5 Qxd5
00:00:02 3.08 -1.28 1.Qf3
00:00:02 3.08 -0.86 1.Qf3 Qh5 2.a5
00:00:02 3.14 -0.80 1.Qd2 h5 2.Nh6 Qxa4
00:00:02 4.00 -0.75 1.Qd2 h5 2.Nh6 Rc8
00:00:02 4.03 -0.75 1.Ne3
00:00:02 4.03 -0.16 1.Ne3 Nxe3 2.Rgxe3 Nd5 3.R3e2
00:00:02 5.00 -0.46 1.Ne3 Nxe3 2.Rgxe3 Nd5 3.Rg3 Kg8
00:00:03 5.02 -0.07 1.Nxh6 Rxh6 2.Nxd6 Qd7 3.Nxb7 Qxb7
4.Qd4 (3)
00:00:03 5.13 -0.04 1.Qd4 Rc8 2.Nxh6 Nf6
00:00:04 6.00 0.26 1.Qd4 Rc8 2.Ne3 Nxe3 3.Rexe3 Qf7
4.Bc3 (3)
00:00:05 7.00 0.26 1.Qd4 Rc8 2.Ne3 Nxe3 3.Rexe3 Qf7
4.Bc3 (5)
00:00:14 8.00 0.43 1.Qd4 Qxa4 2.Ne3 Nxe3 3.Rxg7 Nf5
4.Rh7 Kxh7 (11)
00:00:59 9.00 0.16 1.Qd4 c3 2.Bxc3 Nxc3 3.Qxb4 Nxe4
4.Qxb7 Rb8 5.Qxb8 Qxb8 (24)
00:01:39 9.01 0.50 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4
Kg8 5.Nf6 Rxf6 6.Rxe8 Rxe8 (59)
00:04:11 10.00 0.86 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 Qf7 4.Qh4
Kg8 5.Nh6 Rxh6 (129)
00:10:08 11.00 0.68 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4
Kg8 5.Re2 Qd7 (333)
00:36:41 12.00 0.64 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4
Kg8 5.Rf1 Be7 6.Nxe7 (1150)
02:23:04 13.00 1.00 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4
Kg8 5.Re2 Qd7 (4104)
Positions : 1.230.246.223
Nodes per second : 140.423
Extensions (checks) : 121.164.611 (0%)
Extensions (captures) : 9.758.474 (0%)
Extensions (king safety) : 737.471 (0%)
Extensions (on depth) : 0 (0%)
Extensions (remaining) : 1.150.230 (0%)
Extensions (total) : 132.810.784 (10%)
Hash Table : 7.151.092 6.255.488 (99%)
Average depth : 15
Deepest depth : 60
On Celeron, 500MHz.
Q5T3A: Q5T with Chess Knowledge = 500 which does slow it down a bit, I think
regular Q5T is faster in solving this position, question is still which
variations it has to "see" to regard the position as solved by program..
Jeremiah gave analysis saying that after 3.Qg4 reply 3. ..g6 chosen in Q5T's
main line is not the best. As the evaluation is a !? (By Deep Thought-2?), maybe
that variation could be looked at again also.
--------------------------------------------------------------------------------
Posted by Jeremiah Penery on November 27, 2000 at 12:51:26:
I see Mogens asked for me to post these positions in another thread, but it's
way down near the bottom. I've been wanting to post these for a while anyway,
since it's been a long time since they've been looked at. I'm interested to see
the results from new programs, especially 'speculative' programs like GT, since
these are mostly king-attack positions.
Anyway, here's the original positions, analysis, results from DT-2, and some
other stuff. I've tried to clean it up some, because there were some minor
errors. In #11, I've also inserted what I think is a new winning line for white
in what otherwise would be a drawing line.
One last thing: I don't consider any of these positions truly solved unless the
correct line is found, or a score is found to show that the program understands
the position. Like in #1, it's not solved just by picking Nxh6, but by choosing
the line Nxh6 c3 Nf5!! cxb2 Qg4, which is crushing for white.
--------
Sometime ago, the following 11 positions collected by Pierre Nolot were
posted to rec.games.chess. Pierre claimed that no computers can solve any of
them in reasonable time (and no micros in even a few weeks). A long while back,
he might be right. At this point, a few of them are solvable even under
tournament time control. The ones that we tried overnight are all solvable
in a few hours. DT-2 was searching around 3 million nodes/sec during the
runs. The time probably would be reduced by 5-30 times for the harder
problems if the machine has a better quiescence search.
[D]r3qb1k/1b4p1/p2pr2p/3n4/Pnp1N1N1/6RP/1B3PP1/1B1QR1K1 w - -
Kasparov-Karpov, 20th game 1990
26. Nxh6!! c3
26. ...Rxh6!? 27. Nxd6 Qh5 28. Rg5! Qxd1 29. Nf7+ Kg8 30. Nxh6+
Kh8 31. Rxd1 c3 32. Nf7+ Kg8 33. Bg6! Nf4 34. Bxc3 Nxg6 35. Bxb4
Kxf7 36. Rd7+ Kf6 37. Rxg6+ Kxg6 38. Rxb7 +-
27. Nf5! cxb2 28. Qg4 Bc8
28... g6!? 29. Kh2! Qd7 30. Nh4 Bc6 31. Nc5! dxc5 32. Rxe6 Nf6
33. Nxg6+ Kg7 34. Qg5 Nbd5 35. Ne5 Kh8 36. Nxd7 +-
29. Qh4+ Rh6 30. Nxh6 gxh6 31. Kh2! Qe5 32. Ng5 Qf6 33. Re8 Bf5 34. Qxh6
34. Nf7+ Qxf7 35. Qxh6+ Bh7 36. Rxa8 Nf6 37. Rxf8 Qxf8 38. Qxf8+
Ng8 39. Qg7#
34. ...Qxh6 35. Nf7+ Kh7 36. Bxf5+ Qg6 37. Bxg6+ Kg7 38. Rxa8 Be7 39. Rb8
a5 40. Be4+ Kxf7 41. Bxd5+ +-
Kasparov played Nxh6 here. DT-2 agreed with Kasparov's analysis after
going down the line given in Informant, but could not find the move
on its own in one hour time. When letting it run overnight, it produced
Nxh6 after 8 hours (but would play it with a 6 hours/move time control; that
is, it got first indication that the combination might work after about 6
hours), with the variation 1. Nxh6! c3 2. Nf5 cxb2 3. Qg4 which wins for
white.
(Rest of Jeremiah's post not reproduced here)
--------------------------------------------------------------------------------
Rebel Engine for ECTool. (c) Ed Schröder
Engine version : REBEL CENTURY 2
Rebel personality : Q5T under Century 2.0 in ECTool
Opening book file : RebelC.mvs
Hash table size : 12 MB
Analysis mode : Analyzing next move
Refresh interval : 500 ms
Game begin
00:00 03.00 -0.25 1.Nxh6 Rxh6 2.Nxd6 Qd7 3.Nxc4 Kg8
00:01 05.00 -1.09 1.Ne3 Nxe3 2.Rgxe3 Nd5 3.R3e2 Rc8
00:02 05.01 -0.22 1.Nxh6 c3 2.Ba3 Rc8 3.Bxb4 Nxb4 4.Qd4 Rxh6 5.Qxb4
00:08 06.00 -0.17 1.Nxh6 c3 2.Kh2 Rxh6 3.Nxc3 Qd7 4.Qd4 Nxc3 5.Bxc3
00:22 07.00 -0.19 1.Nxh6 c3 2.Kh2 Rxh6 3.Nxc3 Qh5 4.Qd4 Rf6 5.Qd2
00:24 07.04 -0.14 1.Qd4 Qxa4 2.Qxc4 Qb5 3.Qb3 a5 4.Qa3
00:25 07.05 -0.14 1.Qd2
00:26 07.05 -0.02 1.Qd2 c3 2.Bxc3 Nxc3 3.Qxc3 Qxa4 4.Nh2 Re5
00:30 08.00 -0.24 1.Qd2 Rc8 2.Nxh6 c3 3.Bxc3 Nxc3 4.Rxc3 Rxc3 5.Qxc3 d5
00:33 08.01 -0.24 1.Qd4
00:38 08.01 0.13 1.Qd4 Qd7 2.Qxc4 Qxa4 3.Ra3 Qc6 4.Qb3 Rae8
01:17 09.00 0.30 1.Qd4 Qd7 2.Qxc4 Qxa4 3.Ne3 Qb5 4.Qxb5
03:35 09.02 0.30 1.Nxh6
05:47 09.02 0.40 1.Nxh6 Rxh6 2.Nxd6 Qxe1+ 3.Qxe1 Rxd6 4.Qe2 Rh6 5.Qxc4 Re8
6.Qg4
14:33 10.00 0.64 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4+ Kg8 5.Nh6+ Bxh6
47:56 11.00 0.68 1.Nxh6 c3 2.Rg4 Bc6 3.Nf7+ Qxf7
30:12 12.00 1.65 1.Nxh6 c3 2.Nf5 cxb2 3.Qg4 g6 4.Qh4+ Kg8 5.Nh6+ Bxh6
Eelco.
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.