AlphaZero - Stockfish game 3

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Jouni
Posts: 2002
Joined: Wed Mar 08, 2006 7:15 pm

AlphaZero - Stockfish game 3

Post by Jouni » Mon Dec 11, 2017 8:06 am

SF played -Rf8 losing immediately. My SF thinks it's 10. best move:

[D]4r1kq/p2prp1p/5RpP/2p5/7Q/1B4P1/P4PK1/8 b - - 0 1

Analysis by Stockfish 8 64 BMI2:

1. +/= (0.65): 49...Kf8 50.Qf4 Qg8 51.Qc7 c4 52.Qxc4 Rd8 53.Qd4 Ke8 54.g4 Qf8 55.g5 a5 56.Ra6 Re1 57.Qc3 Re4 58.Kf3 Re6 59.Rxa5 Ke7 60.Bxe6 fxe6+ 61.Kg3 Qh8 62.Qc5+ d6 63.Qb4 Rd7 64.Kg2 Qf8 65.a4 Qc8 66.Qf4 Ke8 67.Qf3 Rf7 68.Qa8 Kd8 69.Ra6 Rd7 70.a5 Qxa8+ 71.Rxa8+ Kc7 72.Ra7+ Kc6 73.Rxd7
2. +- (3.20): 49...c4 50.Bxc4 Kf8 51.Qd4 Qg8 52.Rd6 g5 53.Bb5 a6 54.Bxd7 Rxd7 55.Rxd7 Qg6 56.Qh8+ Qg8 57.Qc3 Qg6 58.Qc5+ Kg8 59.Rd6 f6 60.Qd5+ Qf7 61.Qc6 Qg6 62.Rxf6 Qe4+ 63.Qxe4 Rxe4 64.Rxa6 Kf7 65.Kf3 Rc4 66.a4 g4+ 67.Ke3 Rb4 68.a5 Ra4 69.Ra8 Kg6 70.a6 Kxh6 71.a7 Kg7 72.Kd3 h6 73.Kc2 Ra3 74.Kb2 Ra4 75.Kb3 Ra1 76.Kb4 Ra2 77.Kc4 Kh7 78.Kc5
3. +- (4.26): 49...g5 50.Qxg5+ Kf8 51.Bd5 Qg8 52.Qf4 Re6 53.Bxe6 dxe6 54.Qd6+ Re7 55.Qb8+ Re8 56.Qxa7 Rd8 57.Qxc5+ Ke8 58.a4 Qf8 59.Qc6+ Ke7 60.Rf4 f5 61.Rb4 Kf6 62.Rb6 Qe7 63.Qc3+ Kg6 64.Qe5 Re8 65.a5 Qf6 66.Qe3 f4 67.gxf4 Kxh6 68.a6 Qg6+ 69.Kf1 Qg4 70.f5+ Kh5 71.fxe6 Qd1+ 72.Qe1 Qd4 73.Qe2+ Kg6 74.Rb7 Kh6 75.Rd7 Qa1+ 76.Rd1 Qa4
4. +- (7.36): 49...Re6 50.Bxe6 fxe6 51.Qf4 c4 52.Rf7 Qb2 53.Rg7+ Qxg7 54.hxg7 Kxg7 55.Qxc4 Kf7 56.Qf4+ Ke7 57.Qh4+ Kf7 58.Qxh7+ Kf6 59.Qxd7 Re7 60.Qd4+ Kf7 61.Qf4+ Kg7 62.Qe5+ Kf7 63.g4 Kg8 64.Qf6 Rg7 65.Qxe6+ Kh7 66.Qd5 Re7 67.Qc5 Rf7 68.a4 Rg7 69.a5 Kg8 70.Qe5 Kh7 71.Qd5 Kh6 72.Qf3 Kh7 73.Qh3+ Kg8 74.Qc3 Kh7 75.Qe5 Rf7
5. +- (8.24): 49...Rc8 50.Qf4 Kf8 51.Rxf7+ Ke8 52.Rg7 Kd8 53.Qg5 Qf8 54.Rg8 Qxg8 55.Bxg8 Ke8 56.Qf6 a6 57.Qh8 Kd8 58.Bxh7+ Kc7 59.Bg8 Rf8 60.h7 Re2 61.Qc3 Rfxf2+ 62.Kh3 Rh2+ 63.Kg4 Kc6 64.h8Q Rxh8 65.Qxh8 d6 66.Qf6 Re4+ 67.Kf3 Re8 68.Bb3 Rb8 69.Qxg6 Kc7 70.Qf7+ Kc6 71.Qd5+ Kc7
6. +- (8.31): 49...d5 50.Bxd5 Kf8 51.Qf4 Qg8 52.Qd6 a6 53.Qxc5 Rd8 54.Bb3 g5 55.Qb6 Ra8 56.Qd6 g4 57.Bxf7 Qxf7 58.Rxf7+ Kxf7 59.Qd5+ Re6 60.Qxa8 Rxh6 61.Qd5+ Kg7 62.Qb7+ Kf8 63.Qc8+ Kf7 64.Qxg4 Rf6 65.Qd7+ Kg6 66.Qd5 h6 67.g4 Kh7 68.Qe4+ Kg7 69.f4 Kf7 70.Qe5 Re6 71.Qd5 Ke7 72.Kf3 Rg6 73.Qe5+ Re6 74.Qc7+ Kf8 75.Qb8+ Kf7 76.Qa7+ Ke8 77.Qd4
7. +- (8.53): 49...Rb8 50.Qf4 Rxb3 51.axb3 Re8 52.Rxf7 Qd4 53.Rg7+ Qxg7 54.hxg7 Kxg7 55.Qc7 Kf6 56.Qxc5 a6 57.Qc3+ Ke7 58.Qe3+ Kf8 59.Qf3+ Kg7 60.Qb7 Re6 61.Qxd7+ Kf6 62.Qxh7 Rd6 63.Qc7 Ke6 64.Kf3 Kd5 65.Qc4+ Ke5 66.Kg4 Kf6 67.Qc5 Ke6 68.Qc7 Rd7 69.Qc4+ Ke7 70.Qxa6 Kf7 71.Qc4+ Ke7 72.Qe4+ Kd8 73.Qxg6 Kc7 74.Kf3 Kc8 75.Kg2 Kc7 76.b4
8. +- (9.04): 49...a5 50.Qf4 Re6 51.Rxf7 Qd4 52.Rg7+ Qxg7 53.Bxe6+ dxe6 54.hxg7 Kxg7 55.Qe5+ Kh6 56.Qf6 e5 57.Qf7 Rh8 58.Qe7 Rg8 59.Qxe5 Rf8 60.Qe3+ Kg7 61.Qc3+ Rf6 62.Qxa5 Rf5 63.Qc7+ Kg8 64.Qd8+ Kf7 65.Qd7+ Kf6 66.Qxh7 g5 67.Qh6+ Ke5 68.Qg7+ Kd5 69.Qg6 Re5 70.Qd3+ Ke6 71.a4 g4 72.Qc4+ Kd6 73.Qxg4 Re7 74.a5 Rc7 75.Qg6+ Kd5 76.Qb6 Rd7 77.Qg6 Rd6
9. +- (9.84): 49...Ra8 50.Qf4 Re6 51.Bxe6 dxe6 52.g4 g5 53.Qxg5+ Kf8 54.Qf4 Kg8 55.Rxf7 Qd4 56.Rg7+ Qxg7 57.hxg7 Kxg7 58.Qc7+ Kg8 59.Qxc5 a5 60.Qe7 Rf8 61.Qxe6+ Rf7 62.Qc4 h6 63.f4 Kg7 64.Qd4+ Kh7 65.Qe4+ Kg8 66.Qa8+ Kh7 67.Qxa5 Re7 68.Kf3 Kg7 69.Qa4 Re1 70.Qd4+ Kg8 71.Qc4+ Kh7 72.Qd3+ Kg8 73.Qg6+ Kf8
10. +- (10.29): 49...Rf8 50.Qf4 a5 51.Bd5 c4 52.Bxc4 d5 53.Bxd5 Rd7 54.Bb3 Rc7 55.Qd6 Ra8 56.g4 Rd7 57.Qf4 Rf8 58.g5 Rb7 59.Qd6 Rxb3 60.axb3 Ra8 61.Qd5 Rf8 62.Qc4 Qxf6 63.gxf6 Rd8 64.Qc7 Ra8 65.Qd6 Rc8 66.Qd5 Kf8 67.Qxa5 Ke8 68.Qe5+ Kd7 69.Qd5+ Kc7 70.Qxf7+ Kc6 71.Qxh7 Rf8 72.Qxg6

Why hasn't DeepMind published anything new in their home page about possible "breakthrough" ??
Jouni

CheckersGuy
Posts: 273
Joined: Wed Aug 24, 2016 7:49 pm

Re: AlphaZero - Stockfish game 3

Post by CheckersGuy » Mon Dec 11, 2017 3:47 pm

Demis Hassabis mentioned on his twitter that they will publish the full paper soon. The paper they published was pretty much just a preprint. I guess that they will have additional inforation about A0 on their homePage when they release the full paper

MikeGL
Posts: 900
Joined: Thu Sep 01, 2011 12:49 pm

Re: AlphaZero - Stockfish game 3

Post by MikeGL » Mon Dec 11, 2017 10:33 pm

Jouni wrote:SF played -Rf8 losing immediately. My SF thinks it's 10. best move:

[D]4r1kq/p2prp1p/5RpP/2p5/7Q/1B4P1/P4PK1/8 b - - 0 1

Analysis by Stockfish 8 64 BMI2:

1. +/= (0.65): 49...Kf8 50.Qf4 Qg8 51.Qc7 c4 52.Qxc4 Rd8 53.Qd4 Ke8 54.g4 Qf8 55.g5 a5 56.Ra6 Re1 57.Qc3 Re4 58.Kf3 Re6 59.Rxa5 Ke7 60.Bxe6 fxe6+ 61.Kg3 Qh8 62.Qc5+ d6 63.Qb4 Rd7 64.Kg2 Qf8 65.a4 Qc8 66.Qf4 Ke8 67.Qf3 Rf7 68.Qa8 Kd8 69.Ra6 Rd7 70.a5 Qxa8+ 71.Rxa8+ Kc7 72.Ra7+ Kc6 73.Rxd7
2. +- (3.20): 49...c4 50.Bxc4 Kf8 51.Qd4 Qg8 52.Rd6 g5 53.Bb5 a6 54.Bxd7 Rxd7 55.Rxd7 Qg6 56.Qh8+ Qg8 57.Qc3 Qg6 58.Qc5+ Kg8 59.Rd6 f6 60.Qd5+ Qf7 61.Qc6 Qg6 62.Rxf6 Qe4+ 63.Qxe4 Rxe4 64.Rxa6 Kf7 65.Kf3 Rc4 66.a4 g4+ 67.Ke3 Rb4 68.a5 Ra4 69.Ra8 Kg6 70.a6 Kxh6 71.a7 Kg7 72.Kd3 h6 73.Kc2 Ra3 74.Kb2 Ra4 75.Kb3 Ra1 76.Kb4 Ra2 77.Kc4 Kh7 78.Kc5
3. +- (4.26): 49...g5 50.Qxg5+ Kf8 51.Bd5 Qg8 52.Qf4 Re6 53.Bxe6 dxe6 54.Qd6+ Re7 55.Qb8+ Re8 56.Qxa7 Rd8 57.Qxc5+ Ke8 58.a4 Qf8 59.Qc6+ Ke7 60.Rf4 f5 61.Rb4 Kf6 62.Rb6 Qe7 63.Qc3+ Kg6 64.Qe5 Re8 65.a5 Qf6 66.Qe3 f4 67.gxf4 Kxh6 68.a6 Qg6+ 69.Kf1 Qg4 70.f5+ Kh5 71.fxe6 Qd1+ 72.Qe1 Qd4 73.Qe2+ Kg6 74.Rb7 Kh6 75.Rd7 Qa1+ 76.Rd1 Qa4
4. +- (7.36): 49...Re6 50.Bxe6 fxe6 51.Qf4 c4 52.Rf7 Qb2 53.Rg7+ Qxg7 54.hxg7 Kxg7 55.Qxc4 Kf7 56.Qf4+ Ke7 57.Qh4+ Kf7 58.Qxh7+ Kf6 59.Qxd7 Re7 60.Qd4+ Kf7 61.Qf4+ Kg7 62.Qe5+ Kf7 63.g4 Kg8 64.Qf6 Rg7 65.Qxe6+ Kh7 66.Qd5 Re7 67.Qc5 Rf7 68.a4 Rg7 69.a5 Kg8 70.Qe5 Kh7 71.Qd5 Kh6 72.Qf3 Kh7 73.Qh3+ Kg8 74.Qc3 Kh7 75.Qe5 Rf7
5. +- (8.24): 49...Rc8 50.Qf4 Kf8 51.Rxf7+ Ke8 52.Rg7 Kd8 53.Qg5 Qf8 54.Rg8 Qxg8 55.Bxg8 Ke8 56.Qf6 a6 57.Qh8 Kd8 58.Bxh7+ Kc7 59.Bg8 Rf8 60.h7 Re2 61.Qc3 Rfxf2+ 62.Kh3 Rh2+ 63.Kg4 Kc6 64.h8Q Rxh8 65.Qxh8 d6 66.Qf6 Re4+ 67.Kf3 Re8 68.Bb3 Rb8 69.Qxg6 Kc7 70.Qf7+ Kc6 71.Qd5+ Kc7
6. +- (8.31): 49...d5 50.Bxd5 Kf8 51.Qf4 Qg8 52.Qd6 a6 53.Qxc5 Rd8 54.Bb3 g5 55.Qb6 Ra8 56.Qd6 g4 57.Bxf7 Qxf7 58.Rxf7+ Kxf7 59.Qd5+ Re6 60.Qxa8 Rxh6 61.Qd5+ Kg7 62.Qb7+ Kf8 63.Qc8+ Kf7 64.Qxg4 Rf6 65.Qd7+ Kg6 66.Qd5 h6 67.g4 Kh7 68.Qe4+ Kg7 69.f4 Kf7 70.Qe5 Re6 71.Qd5 Ke7 72.Kf3 Rg6 73.Qe5+ Re6 74.Qc7+ Kf8 75.Qb8+ Kf7 76.Qa7+ Ke8 77.Qd4
7. +- (8.53): 49...Rb8 50.Qf4 Rxb3 51.axb3 Re8 52.Rxf7 Qd4 53.Rg7+ Qxg7 54.hxg7 Kxg7 55.Qc7 Kf6 56.Qxc5 a6 57.Qc3+ Ke7 58.Qe3+ Kf8 59.Qf3+ Kg7 60.Qb7 Re6 61.Qxd7+ Kf6 62.Qxh7 Rd6 63.Qc7 Ke6 64.Kf3 Kd5 65.Qc4+ Ke5 66.Kg4 Kf6 67.Qc5 Ke6 68.Qc7 Rd7 69.Qc4+ Ke7 70.Qxa6 Kf7 71.Qc4+ Ke7 72.Qe4+ Kd8 73.Qxg6 Kc7 74.Kf3 Kc8 75.Kg2 Kc7 76.b4
8. +- (9.04): 49...a5 50.Qf4 Re6 51.Rxf7 Qd4 52.Rg7+ Qxg7 53.Bxe6+ dxe6 54.hxg7 Kxg7 55.Qe5+ Kh6 56.Qf6 e5 57.Qf7 Rh8 58.Qe7 Rg8 59.Qxe5 Rf8 60.Qe3+ Kg7 61.Qc3+ Rf6 62.Qxa5 Rf5 63.Qc7+ Kg8 64.Qd8+ Kf7 65.Qd7+ Kf6 66.Qxh7 g5 67.Qh6+ Ke5 68.Qg7+ Kd5 69.Qg6 Re5 70.Qd3+ Ke6 71.a4 g4 72.Qc4+ Kd6 73.Qxg4 Re7 74.a5 Rc7 75.Qg6+ Kd5 76.Qb6 Rd7 77.Qg6 Rd6
9. +- (9.84): 49...Ra8 50.Qf4 Re6 51.Bxe6 dxe6 52.g4 g5 53.Qxg5+ Kf8 54.Qf4 Kg8 55.Rxf7 Qd4 56.Rg7+ Qxg7 57.hxg7 Kxg7 58.Qc7+ Kg8 59.Qxc5 a5 60.Qe7 Rf8 61.Qxe6+ Rf7 62.Qc4 h6 63.f4 Kg7 64.Qd4+ Kh7 65.Qe4+ Kg8 66.Qa8+ Kh7 67.Qxa5 Re7 68.Kf3 Kg7 69.Qa4 Re1 70.Qd4+ Kg8 71.Qc4+ Kh7 72.Qd3+ Kg8 73.Qg6+ Kf8
10. +- (10.29): 49...Rf8 50.Qf4 a5 51.Bd5 c4 52.Bxc4 d5 53.Bxd5 Rd7 54.Bb3 Rc7 55.Qd6 Ra8 56.g4 Rd7 57.Qf4 Rf8 58.g5 Rb7 59.Qd6 Rxb3 60.axb3 Ra8 61.Qd5 Rf8 62.Qc4 Qxf6 63.gxf6 Rd8 64.Qc7 Ra8 65.Qd6 Rc8 66.Qd5 Kf8 67.Qxa5 Ke8 68.Qe5+ Kd7 69.Qd5+ Kc7 70.Qxf7+ Kc6 71.Qxh7 Rf8 72.Qxg6

Why hasn't DeepMind published anything new in their home page about possible "breakthrough" ??
You are correct, 49...Kf8 seems to hold and is just roughly equal,
while 49.Rf8? is -10.29 in eval. Strange play by SF8 with 49...Rf8 when it
was even said to be on 64 cores. Thanks for the top 10 PV on this position.

MonteCarlo
Posts: 62
Joined: Sun Dec 25, 2016 3:59 pm

Re: AlphaZero - Stockfish game 3

Post by MonteCarlo » Tue Dec 12, 2017 5:59 pm

Yeah, if SF8 is just dumped in this position, on my machine (using just 1 core) it consistently sticks with Rf8 at 0.00 until d35 (takes about 30 seconds on my machine) across a bunch of runs.

However, if I let it play from a couple moves back, so the hash is populated when it arrives at this position, it sticks with Rf8 for MUCH longer.

It then stuck with Rf8 at 0.00 until after 5 billion nodes, finally switching to Kf8 after 53 minutes.

I'm going to play around on some of my other machines once I have access to them again, but after requiring 53 minutes to switch from Rf8 on a single core, it's not at all unlikely that a given run on 64 threads with 1 minute to think could have stuck with Rf8.

Nay Lin Tun
Posts: 519
Joined: Mon Jan 16, 2012 5:34 am

Re: AlphaZero - Stockfish game 3

Post by Nay Lin Tun » Tue Dec 12, 2017 6:16 pm

I am quite sure that even 1300 rated player can realize that this position is hopeless for black. Why do you think that Kf8 can hold /draw??

Spliffjiffer
Posts: 210
Joined: Thu Aug 02, 2012 5:48 pm
Location: Germany

Re: AlphaZero - Stockfish game 3

Post by Spliffjiffer » Tue Dec 12, 2017 7:55 pm

ok, im not asked but i analysed the pos quite deeply some days ago and came to the conclusion that 49...kf8 draws (instead of 49...rf8)...besides that, thanks for clarifying that SF8 has big problems with avoiding 49..rf8....SF.dev has no longer problems with that issue...good news :-))
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.

User avatar
Eelco de Groot
Posts: 4156
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: AlphaZero - Stockfish game 3

Post by Eelco de Groot » Tue Dec 12, 2017 8:17 pm

After 49...Rf8

5rkq/p2prp1p/5RpP/2p5/7Q/1B4P1/P4PK1/8 w - -

Engine: Kaissa (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

21/36 0:01 +1.57 50.Qf4 a6 51.Bd5 c4 52.Bxc4 d5
53.Bxd5 Rd7 54.Bb3 Rc7 55.g4 Rd7
56.a4 Rb7 57.Bc4 Rc7 58.g5 Rd7
59.Bxa6 Rd1 60.Bc4 Rd7 61.a5 Re7 (2.788.740) 1890

22/34 0:02 +2.00 50.Qf4 a6 51.Bd5 c4 52.Bxc4 d5
53.Bxd5 Rd7 54.Bb3 Rc7 55.g4 Rd7
56.a4 a5 57.g5 Rb7 58.Ba2 Rd7 59.Qc4 Rb7
60.Qd5 Ra7 61.Qc5 Rd7 62.Bc4 Rb7 (5.315.509) 1909

23/29 0:03 +1.83-- 50.Qf4 c4 (6.009.979) 1904

23/29 0:03 +2.00++ 50.Qf4 (6.082.281) 1903

23/29 0:03 +2.12++ 50.Qf4 (6.658.507) 1915

23/33 0:03 +2.27++ 50.Qf4 (7.307.034) 1927

23/33 0:04 +2.29 50.Qf4 c4 51.Bxc4 d5 52.Bxd5 Rd7
53.Bc4 Rb7 54.g4 Rd7 55.a4 a5 56.Ba2 Ra7
57.Qd6 Re8 58.g5 Rae7 59.Bd5 Ra7
60.Bb3 Rae7 61.Bc4 Ra7 62.Qd5 Rf8 (7.844.045) 1926

24/32 0:04 +2.37++ 50.Qf4 (8.887.298) 1937

24/32 0:04 +2.44++ 50.Qf4 (9.620.981) 1940

24/35 0:05 +2.54 50.Qf4 c4 51.Bxc4 d5 52.Bxd5 Rd7
53.Bc4 Rb7 54.g4 Rd7 55.a4 a5 56.Ba2 Ra7
57.Qd6 Re8 58.g5 Rae7 59.Bd5 Ra7
60.Bb3 Rae7 61.Bc4 Ra7 62.Bd5 Rae7
63.Ba2 (10.663.174) 1950

25/35 0:05 +2.33-- 50.Qf4 d5 (11.278.258) 1957

25/35 0:05 +2.54++ 50.Qf4 (11.545.160) 1959

25/38 0:06 +2.37-- 50.Qf4 c4 (11.941.224) 1959

25/38 0:06 +2.51++ 50.Qf4 (11.996.787) 1958

25/38 0:06 +2.73++ 50.Qf4 (13.422.231) 1962

25/38 0:07 +3.02++ 50.Qf4 (14.935.856) 1962

25/41 0:08 +3.40++ 50.Qf4 (16.533.332) 1965

25/41 0:09 +3.65 50.Qf4 a6 51.Bd5 a5 52.g4 c4 53.Bxc4 d5
54.Bxd5 Rd7 55.Bb3 a4 56.Bxa4 Re7
57.Bb3 Rd7 58.Bc4 Ra7 59.g5 Rd7
60.a4 Ra8 61.a5 Rb7 62.Bxf7+ Rxf7
63.Rxf7 (19.342.367) 1972

26/34 0:10 +3.73++ 50.Qf4 (20.125.097) 1975

26/34 0:10 +3.80++ 50.Qf4 (20.664.266) 1973

26/39 0:11 +3.92++ 50.Qf4 (22.458.943) 1976

26/47 0:12 +4.07++ 50.Qf4 (24.620.064) 1975

26/55 0:13 +4.29++ 50.Qf4 (26.884.772) 1977

26/55 0:14 +4.58++ 50.Qf4 (29.188.997) 1982

26/55 0:16 +4.97++ 50.Qf4 (32.459.782) 1990

26/55 0:18 +5.47++ 50.Qf4 (36.295.185) 2003

26/59 0:20 +6.11++ 50.Qf4 (41.084.209) 2013

26/59 0:22 +5.72 50.Qf4 d5 51.Bxd5 Rd7 52.Bc4 Rc7
53.Qd6 Ra8 54.Bd5 Re8 55.g4 a6 56.g5 Ra7
57.Qc6 Rb8 58.Rxg6+ hxg6 59.Qxg6+ Kf8
60.Qd6+ Re7 61.Qxb8+ Re8 62.Qd6+ Re7
63.Bc6 (46.210.229) 2035

27/37 0:22 +5.80++ 50.Qf4 (46.807.683) 2037

27/40 0:23 +5.87++ 50.Qf4 (47.744.957) 2039

27/40 0:23 +5.99++ 50.Qf4 (48.365.289) 2042

27/51 0:25 +6.15++ 50.Qf4 (51.535.576) 2054

27/51 0:26 +6.36++ 50.Qf4 (54.872.557) 2064

27/60 0:28 +6.65++ 50.Qf4 (60.308.542) 2083

27/60 0:34 +7.04++ 50.Qf4 (71.838.616) 2105

27/60 0:39 +7.54++ 50.Qf4 (83.702.135) 2135

27/60 0:45 +7.65 50.Qf4 d5 51.Bxd5 Rd7 52.Bc4 a6
53.g4 Ra7 54.Qd6 Re8 55.g5 Rae7
56.Qxc5 Qxf6 57.gxf6 Rd7 58.Bxa6 Ra8
59.Bc4 Rf8 60.a4 Rdd8 61.a5 Kh8
62.a6 Rc8 63.Qe7 (98.827.065) 2158

28/63 0:49 +7.52-- 50.Qf4 c4 (106.955.466) 2166

28/63 0:49 +7.65++ 50.Qf4 (107.689.133) 2167

28/63 0:54 +7.76++ 50.Qf4 (118.825.174) 2179

28/63 0:59 +7.92++ 50.Qf4 (131.392.297) 2193

28/63 1:06 +8.13++ 50.Qf4 (146.459.564) 2214

28/63 1:11 +8.42++ 50.Qf4 (159.224.513) 2227

28/63 1:18 +8.81++ 50.Qf4 (175.495.820) 2241

28/63 1:25 +9.31++ 50.Qf4 (193.348.407) 2256

28/63 1:27 +9.31 50.Qf4 c4 51.Bxc4 d5 52.Bxd5 Rd7
53.Bb3 Rb7 54.Qd6 Re8 55.Bd5 Rc7
56.Qf4 Rf8 57.g4 Rd7 58.Bb3 Rb7
59.Qd6 Rc8 60.Bd5 Rbc7 61.g5 Qxf6
62.Qxf6 Kf8 63.Be6 (196.489.499) 2256

29/51 1:36 +9.38++ 50.Qf4 (218.506.853) 2274

29/54 1:42 +9.46++ 50.Qf4 (235.387.378) 2289

29/54 1:49 +9.57++ 50.Qf4 (251.315.785) 2300

29/54 1:56 +9.73++ 50.Qf4 (267.978.762) 2309

29/55 2:03 +9.95++ 50.Qf4 (287.628.870) 2320

29/60 2:17 +10.24++ 50.Qf4 (323.138.985) 2344

29/60 2:20 +10.24 50.Qf4 c4 51.Bxc4 d5 52.Bxd5 Rd7
53.Bb3 Rb7 54.Qd6 Rc8 55.Bd5 Rbc7
56.g4 g5 57.Qe5 a5 58.Qf5 Qxf6
59.Qxf6 Kf8 60.Qg7+ Ke7 61.Qxh7 Rf8
62.Qe4+ Kd6 63.Qd4 (329.654.803) 2342

30/60 2:21 +10.24 50.Qf4 c4 51.Bxc4 d5 52.Bxd5 Rd7
53.Bb3 Rb7 54.Qd6 Rc8 55.Bd5 Rbc7
56.g4 g5 57.Qe5 a5 58.Qf5 Qxf6
59.Qxf6 Kf8 60.Qg7+ Ke7 61.Qxh7 Rf8
62.Qe4+ Kd6 63.Qd4 (332.295.454) 2345

best move: Qh4-f4 time: 2:21.829 min n/s: 2.345.228 nodes: 332.295.454

If anybody has an idea why Sf8 does not see this, please post that would be helpful. Of course we can go through all the commits since then, one by one, but at the moment I have no idea :?
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

User avatar
Eelco de Groot
Posts: 4156
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: AlphaZero - Stockfish game 3

Post by Eelco de Groot » Tue Dec 12, 2017 8:35 pm

Stockfish 8 is from Nov 1 2016 and this later version does not have the same problem anymore (assuming I did not change anything from the default but I think this is the default 2017 Jan 5th Stock):


5rkq/p2prp1p/5RpP/2p5/7Q/1B4P1/P4PK1/8 w - -

Engine: Stockfish20170105Def (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

31/27 0:01 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (3.592.602) 1801

32/27 0:02 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (4.881.636) 1820

33/27 0:03 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (6.238.240) 1849

34/27 0:04 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (8.662.548) 1882

35/27 0:05 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (10.993.501) 1899

36/27 0:07 0.00 50.Qf4 d5 51.Bxd5 Rd7 52.Bb3 a5
53.Bxf7+ Rdxf7 54.Rxf7 Rxf7 55.Qb8+ Rf8
56.Qb3+ Rf7 57.Qb8+ (13.965.915) 1934

37/36 0:10 0.00 50.Qf4 a5 51.Bxf7+ Rexf7 52.Rxf7 Rxf7
53.Qb8+ Rf8 54.Qb3+ c4 55.Qxc4+ Rf7
56.Qc8+ Rf8 57.Qc4+ (19.658.924) 1924

38/41 0:12 +0.07++ 50.Qf4 (23.332.830) 1921

38/44 0:14 +0.14++ 50.Qf4 (27.184.381) 1930

38/44 0:16 +0.25++ 50.Qf4 (31.262.697) 1941

38/46 0:17 +0.40++ 50.Qf4 (34.745.202) 1945

38/46 0:20 +0.61++ 50.Qf4 (40.293.238) 1959

38/48 0:22 +0.89++ 50.Qf4 (44.723.062) 1965

38/57 0:26 +1.27++ 50.Qf4 (51.215.999) 1963

38/57 0:31 +1.75++ 50.Qf4 (61.898.312) 1965

38/57 0:36 +2.37++ 50.Qf4 (71.040.169) 1961

38/57 0:41 +3.17++ 50.Qf4 (81.377.352) 1954

38/67 0:49 +4.19++ 50.Qf4 (96.473.641) 1953

38/67 1:08 +5.48++ 50.Qf4 (134.261.392) 1960

38/67 1:33 +4.02-- 50.Qf4 c4 (189.260.366) 2013

38/67 1:36 +5.76++ 50.Qf4 (194.435.121) 2018

38/67 1:56 +6.82 50.Qf4 a6 51.Bc4 d5 52.Bxd5 Rd7
53.Bb3 Rc7 54.Bc4 a5 55.g4 Rd7
56.Bb3 Rc7 57.g5 Rd7 58.Qa4 Qxf6
59.gxf6 Ra7 60.Qc6 c4 61.Qb6 Rfa8
62.Bxc4 Rd7 63.Qc6 (238.874.064) 2046

39/67 1:59 +6.72-- 50.Qf4 d5 (244.544.046) 2045

39/67 1:59 +6.72 50.Qf4 d5 (244.646.846) 2045

best move: Qh4-f4 time: 1:59.719 min n/s: 2.045.730 nodes: 244.646.846

Probably this patch from December 12th clears it up, but I have not checked that: Threefold repetition detection
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

MonteCarlo
Posts: 62
Joined: Sun Dec 25, 2016 3:59 pm

Re: AlphaZero - Stockfish game 3

Post by MonteCarlo » Wed Dec 13, 2017 2:07 pm

Yeah, even SF8 doesn't have too hard of a time finding the win after Rf8 is played if you just start the engine on that position (finds it at iteration 38 just like in your posted example).

FEN: 5rkq/p2prp1p/5RpP/2p5/7Q/1B4P1/P4PK1/8 w - - 0 1

Stockfish_8_x64:

25/47 00:02 4,354k 1,882k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
26/47 00:03 5,738k 1,890k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
27/47 00:03 5,881k 1,886k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
27/47 00:03 6,091k 1,882k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
27/47 00:03 7,266k 1,897k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
28/47 00:03 7,401k 1,897k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
28/47 00:04 7,723k 1,890k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
28/47 00:04 8,890k 1,892k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
29/47 00:04 9,289k 1,896k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
29/47 00:05 9,602k 1,894k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
29/50 00:06 11,506k 1,909k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
30/50 00:06 11,806k 1,909k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
30/50 00:06 12,375k 1,905k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
30/50 00:08 16,366k 1,937k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
31/50 00:08 17,019k 1,937k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
31/50 00:09 17,517k 1,933k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
31/50 00:10 20,076k 1,938k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
32/50 00:10 20,859k 1,937k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
32/50 00:10 21,271k 1,935k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
32/50 00:12 23,766k 1,937k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
33/50 00:12 24,438k 1,934k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
33/50 00:12 25,012k 1,932k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
33/50 00:15 29,930k 1,944k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
34/50 00:15 30,602k 1,944k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
34/50 00:16 31,646k 1,942k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
34/50 00:19 38,825k 1,960k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
35/50 00:20 40,088k 1,957k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
35/50 00:21 41,113k 1,956k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
35/50 00:24 48,091k 1,968k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
36/50 00:25 49,400k 1,966k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
36/50 00:27 54,203k 1,974k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
36/50 00:32 63,828k 1,985k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
37/50 00:34 67,862k 1,986k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
37/50 00:35 69,582k 1,982k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
37/50 00:39 79,129k 1,987k 0.00 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 a7-a5 a2-a4 Rd7-b7 Bb3-d5 Rb7-d7 Bd5-b3
38/68 01:23 165,565k 1,988k +2.39 Qh4-f4 d7-d5 Bb3xd5 Re7-d7 Bd5-b3 Rd7-c7 g3-g4 c5-c4 g4-g5 Rf8-c8 Bb3-a4 Rc7-c5 Ba4-d7 Rc8-d8 Qf4-e3 Rc5-c7 Rf6xf7 Kg8xf7 Qe3-e6+ Kf7-f8 Qe6-d6+ Kf8-f7 Bd7-e6+ Kf7-e8 Qd6xc7 Qh8-f8 Qc7xh7 Qf8-f4 Qh7xg6+ Ke8-e7 Qg6-f5 Qf4xf5 Be6xf5 Rd8-d5 g5-g6 Rd5-d8 g6-g7 Ke7-f6 Kg2-f3 c4-c3 Bf5-c2 a7-a5 a2-a4 Kf6-f7 Kf3-g4 Kf7-g8 Bc2-b3+ Kg8-h7 Kg4-g5 Rd8-b8 Bb3-c2+ Kh7-g8 f2-f4 Kg8-f7 Kg5-g4 Kf7-g8
38/69 02:23 284,741k 1,991k +4.85 Qh4-g5 Rf8-c8 Qg5-f4 Rc8-f8 Bb3-d5 a7-a6 a2-a4 c5-c4 Bd5xc4 d7-d5 Bc4xd5 Re7-d7 Bd5-c4 Rd7-c7 Rf6xa6 Rc7-d7 Ra6-f6 Rd7-c7 Rf6-b6 Rc7-c8 Rb6-b7 Qh8-c3 Rb7xf7 Rf8xf7 Qf4xf7+ Kg8-h8 a4-a5 Qc3-d4 a5-a6 Qd4-e4+ Kg2-h2 Qe4-e5 a6-a7 Qe5-h5+ Kh2-g2 Qh5xh6 Bc4-d5 Qh6-f8 Qf7-b7 Rc8-d8 Qb7-b2+ Qf8-g7 Qb2-a2 Qg7-d4 a7-a8Q Rd8xa8 Qa2xa8+ Kh8-g7 Qa8-b7+ Kg7-h6 Qb7-d7 Qd4-d2 Qd7-f7 Qd2-b2 Qf7-e6 Kh6-g7 Qe6-e7+ Kg7-h6 Qe7-h4+ Kh6-g7 Qh4-d8
38/69 02:27 295,330k 1,997k +4.85 Qh4-g5 Rf8-c8 Qg5-f4 Rc8-f8 Bb3-d5 a7-a6 a2-a4 c5-c4 Bd5xc4 d7-d5 Bc4xd5 Re7-d7 Bd5-c4 Rd7-c7 Rf6xa6 Rc7-d7 Ra6-f6 Rd7-c7 Rf6-b6 Rc7-c8 Rb6-b7 Qh8-c3 Rb7xf7 Rf8xf7 Qf4xf7+ Kg8-h8 a4-a5 Qc3-d4 a5-a6 Qd4-e4+ Kg2-h2 Qe4-e5 a6-a7 Qe5-h5+ Kh2-g2 Qh5xh6 Bc4-d5 Qh6-f8 Qf7-b7 Rc8-d8 Qb7-b2+ Qf8-g7 Qb2-a2 Qg7-d4 a7-a8Q Rd8xa8 Qa2xa8+ Kh8-g7 Qa8-b7+ Kg7-h6 Qb7-d7 Qd4-d2 Qd7-f7 Qd2-b2 Qf7-e6 Kh6-g7 Qe6-e7+ Kg7-h6 Qe7-h4+ Kh6-g7 Qh4-d8


Of course, as I pointed out previously, it also doesn't have too hard of a time avoiding Rf8 if you just start it immediately on that position.

The big problem seems to be if you approach the position from a few moves back (I did a couple minutes per move) so that the hash is already populated when you get there. Then it takes much, much longer to get over Rf8 (well, to be fair, it took much, much longer the one time I ran that test; it just took so long I haven't tried reproducing it yet, but I will eventually).

Jouni
Posts: 2002
Joined: Wed Mar 08, 2006 7:15 pm

Re: AlphaZero - Stockfish game 3

Post by Jouni » Wed Dec 13, 2017 8:09 pm

This move by SF8 was obviously caused by buggy repetion code. I got it sometimes:

Analysis by Stockfish 8 64 BMI2:

49...Rf8 50.Qf4 a6 51.Qf3 d6 52.Bxf7+ Rexf7 53.Rxf7 Rxf7 54.Qa8+ Rf8 55.Qd5+ Rf7 56.Qa8+
= (0.00) Depth: 43/34 00:00:07 63337kN, tb=16441

SFDev is bettered in this now. Another reason to use latest version!
Jouni

Post Reply