Win or draw?

[bob]

Hamsters can also find the draw:

Code: Select all

Hamsters 0.5 by Alessandro Scotti
a b2rk3/Q7/p3p3/P6p/2Pp3P/8/6P1/6K1 b - - 0 1
 2/ 6 0:00.00 -171 123 Be4 Qxa6
 3/ 8 0:00.00 -172 405 Be4 Qxa6 Ke7
 4/ 9 0:00.00 -168 968 Be4 Qxa6 Ke7 Qa7+ Kf6
 5/11 0:00.00 -176 2751 Be4 Qxa6 Ke7 c5 e5
 6/15 0:00.01 -214 11104 Be4 Qxa6 Bf5 Qb5+ Ke7 a6 Kf6
 7/17 0:00.03 -216 20508 Be4 Qxa6 Bf5 Qb5+ Ke7 a6 Kf6 a7 Rg8
 8/19 0:00.08 -217 49051 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 a7 d2 Qc7+ Rd7 Qc5+ Kf6 a8=Q d1=Q+ Kf2
 9/21 0:00.15 -221 91661 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Ra7 Qxd4 Rxa5
10/25 0:00.41 -219 241736 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Ke8 a6 Kf7 c5
11/27 0:00.81 -218 479222 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Bg4 a6 d3 Qg5+ Kf7 c5
12/27 0:01.64 -213 976629 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Bg4 a6 d3 Qd2 e5 c5
13/30 0:02.61 -219 1555666 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Bg4 a6 d3 Qd2 e5 c5 Ke6
14/31 0:04.47 -227 2670178 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qf4 Bg4 a6 d3 Qg5+ Kd6 Qe3 Ke7 a7 d2 a8=Q d1=Q+ Kh2
15/38 0:10.24 -218 6028089 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Kf7 a6 d3 a7 d2 a8=Q d1=Q+ Kh2 Rd8 Qaa7+ Qd7 Qac7 Qxc7+ Qxc7+ Ke8 c5
16/41 0:19.51 -222 11443650 Be4 Qxa6 Bf5 Qb5+ Kf7 Qb7+ Ke8 Qf3 Bg4 Qc6+ Ke7 Qc5+ Ke8 a6 d3 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Kf6 a7 d2 Qg5+ Kf7 a8=Q d1=Q+ Kh2
17/42 0:37.80 -221 22148776 Be4 Qxa6 Bf5 Qb5+ Kf7 Qb7+ Ke8 Qf3 Bg4 Qc6+ Ke7 Qc5+ Ke8 Qb5+ Kf7 Qg5 Rd7 a6 d3 Qd2 Ke7 c5
17/42 1:28.18 -167 51509570 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc8 Qd2 Be4 g3 e5 Kf2 Rf7+ Ke3 Bc6 Qh2 Rf3+ Kd2 e4 a6 Rf7 Qh3+ Kb8
18/46 3:32.59 0 122136311 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qd1 Be4 a6 Bc2 Qe2+ Kf6 Qf2+ Bf5 Qb6+ Ke7 Qc7+ Rd7 Qe5+ Kd8 Qa5+ Ke7 Qe5+
19/51 5:18.26 0 181466597 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qe5+ Kf7 Qf5+ Ke7 Qh7+ Kf6 Qh6+ Kf7 Qf4+ Ke8 Qe5+ Kf7

FInding such a draw doesn't always mean much. It could easily mean that the engine knows what is going on. Or it has no idea. To reach a repetition, one side has to think it is behind and therefore the repetition is better than a - score. If the wrong side thinks it is losing, then repetitions are far easier to find than if that side was playing for a win and the other side had to try to escape with the repetition draw.

I've seen half my "I found the draw" tests end badly for this very reason. I therefore am not too excited when a program sees the draw until it plays against an opponent that is strong to see if he can find a winning line from either side that the test program could not... It is common that one can...

[Alessandro Scotti]

FInding such a draw doesn't always mean much.

Aww I was so happy and now you've spoiled it!

[Eelco de Groot]

FInding such a draw doesn't always mean much.

Aww I was so happy and now you've spoiled it!

Well I think it was a good call from Hamsters here!

Compare it also to the times for Glaurung Lodz 2007 Default, 41 minutes

00:00:00.5 -1,43 4 1666 Be4 Qxa6 Ke7 Qb5
00:00:00.5 -1,98 5 3227 Be4 Qxa6 Ke7 Qa7+ Kf6 Kf2
00:00:00.5 -1,72 5 6613 e5 Qc5 e4 Qxh5+ Ke7 Qh7+ Kd6 Qg6+ Kc5 Qxa6
00:00:00.5 -2,41 6 9379 e5 Qc5 e4 Qxh5+ Ke7 Qh7+ Kd6 Qg6+ Ke7 Qxa6
00:00:00.5 -2,21 6 11779 Be4 Qxa6 Ke7 Qb5 Bd3 a6
00:00:00.6 -2,47 7 23558 Be4 Qxa6 Bf5 Qb5+ Ke7 Kf2 d3 Ke3
00:00:00.6 -2,43 8 38938 Be4 Qxa6 Bf5 Qb5+ Kf7 Kf2 d3 Qb7+ Kf6 Ke3
00:00:00.7 -2,45 9 77260 Be4 Qxa6 Bf5 Kf2 d3 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Kf6 Ke3
00:00:00.8 -2,39 10 177777 Be4 Qxa6 Bf5 Qb5+ Ke7 Kf2 d3 Qb4+ Kf6 Qb2+ Ke7 Ke3 Rg8 Qb4+ Kd7
00:00:01.2 -2,47 11 459124 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 Qc7+ Rd7 Qc5+ Kf7 Qb5 Ke7 Kf2
00:00:01.9 -2,35 12 924139 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 Qc7+ Rd7 Qc5+ Kf7 Qb5 Rd8 Kf2 d2
00:00:02.8 -2,37 13 1587924 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 Qc7+ Rd7 Qc5+ Kf7 Qb5 Rd8 Qb6 Ke7 Kf2
00:00:04.8 -2,54 14 3008772 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 Qc7+ Rd7 Qc5+ Kf7 a7 d2 a8Q d1Q+ Kh2 Rd6 Qb7+ Rd7 Qbc6
00:00:11.7 -2,31 15 8073489 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc5+ Ke8 Qb5+ Ke7 Qb4+ Ke8 Kf2 d3 Qd2 Rc8 Qg5 Kd7 Ke3 Kd6 Qxh5 Rxc4
00:00:19.4 -2,31 16 13845830 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc5+ Ke8 Qb5+ Ke7 Qb4+ Ke8 Kf2 d3 Qd2 Rc8 Qg5 Rxc4 Qxh5+ Kd7 Ke3 Kd6
00:00:30.7 -2,68 17 22258409 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Rd6 Qb4 d3 c5 Rd5 c6+ Kf6 Qf8+ Ke5 c7 d2 c8Q d1Q+ Kh2 Rd3 Qcc5+ Ke4
00:00:58.3 -2,41 17 42620812 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc7 Qd2 Bc6 Kf2 e5 c5 Bb5 Ke3 Rd5 Qd1 d2 Ke4
00:01:40.6 -2,23 18 74315700 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc8 Qd2 Bb7 Kf2 e5 c5 Rd4 Ke1 Bc6 a6 Kd7 a7 e4
00:02:47.1 -2,11 19 122879937 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc8 Qd2 Bb7 Kf2 e5 Ke1 Bc6 Kf2 Kc7 a6 Kb6 a7 Kxa7 Qa5+ Kb7 Qxe5
00:04:57.6 -1,60 20 218791211 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Rd6 Qg5+ Kd7 Qg7+ Kc6 Qf8 Kc5 Qb8 d2 Qb5+ Kd4 Qxh5 Be4 c5 Rd5 Qd1 Kxc5 Kf2 Kb5 g4 Kxa5
00:08:57.1 -1,54 21 396816349 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Rd6 Qg5+ Kd7 Qg7+ Kc6 Qf8 Kc5 Qb8 d2 Qb5+ Kd4 Qxh5 Kxc4 Qd1 Kb5 Kf2 Be4 Qb3+ Kc5 Qa3+ Kc6 Qa4+ Kc5 Qd1
00:16:25.0 -1,39 22 727710529 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Rd6 Qg5+ Kd7 Qg7+ Kc6 Qg8 Kc5 Qg5+ Kxc4 Qxh5 d2 Qd1 Kb5 a6 Kxa6 Kf2 Bc6 g4 Kb5 Ke3 Kc4 Qc2+ Kb4
00:41:00.8 0,00 23 1812162967 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qg5+ Ke8 Qb5+ Ke7 Qc5+ Ke8 Qe3+ Kf7 Qf4+ Ke8 Qe5+ Kf8 Qf5+ Ke7 Qh7+ Kf6 Qh6+ Kf7 Qh5+ Ke7
00:54:32.3 0,00 24 2409091811 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qg5+ Ke8 Qb5+ Ke7 Qc5+ Ke8 Qe3+ Kf7 Qf4+ Ke8 Qe5+ Kf8 Qf5+ Ke7 Qh7+ Kf6 Qh6+ Kf7 Qh5+ Ke7

Glaurung 2.0 Beta 071001 Default here in 23 minutes

00:00:00.3 -2,05 4 1351 e5 Kf2 e4 Qxa6 e3+ Kg3
00:00:00.3 -2,21 5 4720 e5 Qc5 e4 Qxh5+ Kd7 Qf7+ Kc6 Qe6+ Kc5 Qxa6
00:00:00.3 -2,86 6 8973 e5 Qc5 e4 Qxh5+ Ke7 Qh7+ Ke6 Qg6+ Kd7 Qxa6
00:00:00.3 -2,82 6 12200 Be4 Qxa6 Ke7 Qb5 Bd3 a6
00:00:00.4 -2,76 7 23120 Be4 Qxa6 Bf5 Kf2 d3 Qb7 d2
00:00:00.4 -2,82 8 33543 Be4 Qxa6 Bf5 Qb5+ Ke7 Qc5+ Kf6 Qb6 Ke7 Kf2
00:00:00.5 -2,80 9 73380 Be4 Qxa6 Bf5 Qb5+ Ke7 Qb7+ Kf6 Qf3 d3 Qxh5 Ke5
00:00:00.5 -2,88 10 150894 Be4 Qxa6 Bf5 Qc6+ Ke7 a6 d3 a7 d2 Qc7+ Rd7 Qc5+ Kf6 a8Q d1Q+ Kf2
00:00:00.8 -2,92 11 283866 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qg3 d3 Qg5+ Kd6 Qxh5 Rh7 Qg5
00:00:01.4 -2,86 12 677216 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Rd6 a6 d3 a7 d2 Kf2 d1Q a8Q Qd4+ Qxd4 Rxd4
00:00:02.2 -3,09 13 1246884 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Rd6 a6 d3 a7 d2 a8Q d1Q+ Kh2 Kf7 Qe5
00:00:04.6 -2,98 14 2932999 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Rd6 a6 d3 a7 d2 a8Q d1Q+ Kh2 Kf7 Qca7+ Rd7 Qe3
00:00:07.0 -3,00 15 4739492 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Rd6 Qe5 Kd7 Qg7+ Kc6 Qh8 Kc5 Qxh5 Kxc4 g4 Be4
00:00:15.5 -2,90 16 10828861 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc5+ Ke8 Qb5+ Ke7 Qb4+ Ke8 Qb2 Kf7 Qb7+ Kf6 Qb6 Rd7 a6 d3 Qb2+ Kf7 Kf2
00:00:26.8 -2,94 17 19000564 Be4 Qxa6 Bf5 Qc6+ Ke7 Qc5+ Ke8 Qb5+ Ke7 a6 d3 Qc5+ Ke8 Qc6+ Ke7 Qc7+ Rd7 Qc5+ Kf7 a7 d2 a8Q d1Q+ Kh2 Qd6+ Qxd6 Rxd6 Kg3 Rd3+ Kf4
00:00:54.2 -3,13 18 38774277 Be4 Qxa6 Bf5 Qb5+ Ke7 Qb4+ Ke8 Kf2 d3 Qd2 Kd7 a6 Rb8 Ke3 Kd6 a7 Ra8 Qb4+ Kc7 Qe7+ Kc6 Kd4
00:01:54.8 -3,00 18 82303074 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc7 Qd2 Bc6 Kf2 e5 a6 Rd4 a7 Rf4+ Ke1 Re4+ Kd1 Rxc4 Qxd3 Rxh4
00:02:44.3 -2,98 19 118358467 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc7 Qd2 Kc8 Kf2 Bb7 Ke3 Ba6 c5 Rd5 Qc1 d2 Qd1 Kd7 g3
00:04:41.0 -2,92 20 203789853 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Kd8 Qg5+ Kc7 Qd2 Bb7 Kf2 e5 c5 Kc6 a6 Bxa6 Qh6+ Kxc5 Qxa6 d2 Qc8+ Kd6 Ke2 Ke7
00:07:52.3 -1,98 21 342837958 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Rd6 Qg5+ Kd7 Qg7+ Kc6 Qg8 Kc5 Qg5+ Kxc4 Qxh5 d2 Qd1 Kb5 a6 Kxa6 Kf2 Kb5 g4 Bb7 Ke3
00:11:21.7 -1,72 22 495112896 d3 Qxa6 Ke7 Qa7+ Rd7 Qc5+ Rd6 Qg5+ Kd7 Qg7+ Kc6 Qg8 Kc5 Qg5+ Kxc4 Qxh5 d2 Qd1 Kb5 a6 Kxa6 Kf2 Kb5 g4 Bb7 Ke3 Kc4 Qc2+ Kb4
00:23:01.9 0,00 23 998610335 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qd1 Be4 Kf2 Bc2 Qe2+ Kf6 Qf3+ Bf5 Qd1 Bc2 Qf3+
00:35:05.1 0,00 24 1514641390 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qd1 Be4 Kf2 Bc2 Qe2+ Kf6 Qf3+ Bf5 Qd1 Bc2 Qf3+
00:53:54.9 0,00 25 2326670465 d3 Qxa6 d2 Qxe6+ Kf8 Qf5+ Ke7 Qg5+ Ke8 Qxh5+ Ke7 Qd1 Be4 Kf2 Bc2 Qe2+ Kf6 Qf3+ Bf5 Qd1 Bc2 Qf3+

About the sinificance of finding a draw score I can understand that finding a draw score is not the same as actually finding the correct drawing line but is that not very similar then as finding a winning combination by the wrong continuation, or the wrong line at first and therefore a high score that is not really coorrect for the PV. The difference then is probably that there are usually more possible repetitions (and some of them not correct) in a eternal check situation or similar, than there are misleading winning lines in a combination position. Maybe I see it wrong. But here it seems there are at least two key moves, d3 followed by d2 that are necessary for the draw, and both Glaurung and Hamsters play this. I must confess I have not analyzed the position further but the PV seem to remain constant at further iterations. I think it has to be said that a program that finds the difficult draw often, probably gains much more elopoints from this than a program that can find an equally difficult mate but has to search for this in many many positions where there is not ever mate to be found?

Eelco

[Eelco de Groot]

Sorry I meant to type:
About the significance of finding a draw score; I can understand that finding a draw score is not the same as actually finding the correct drawing line, but is that not very similar then to finding a winning combination by the wrong continuation, or the wrong line at first and therefore a high score that is not really correct for the found PV?

Eelco

[bob]

My point is that there are two sides in the game. One is really winning, and one is really losing, in these kinds of positions. The winning side wants to win, and the losing side would love to draw.

In a game position, played by a computer, it will follow the above exactly, except that its idea of who is winning or losing is not always correct. And the search will lock onto the draw if it believes it is losing even by a very tiny margin. And in some cases it believes that both sides will play toward a draw since any other move by either side leads to what it believes is a worse score for that side.

Seeing a draw for the wrong reason is really a pretty common thing. Mates are forced onto the side that is getting mated. But draws are not, which makes trusting those scores much harder.