AlphaZero SF game 10 Does Your Engine Find 29Qh3?

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.
FWCC
Posts: 117
Joined: Wed Aug 22, 2007 2:39 pm

AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by FWCC » Sun Dec 24, 2017 2:49 am

rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29
In this position will your engine find Qh3 which engines say is STILL EQUAL?
I used latest SF and the move was not a choice.Engines like 29Bd4+

This game is amazing to me due to this position:
rn3r2/p3b2p/2p3k1/1p3bp1/8/q1P1B1P1/P4PB1/3RR1KQ b - - 0 26
The Queen is positioned on H1 out of play.This looks like something from science fiction,an Alien game!After going over this game,the whole point was that Black's Queenside was NEVER developed,and here AlphaZero makes the TAL decision to hang the knight for unseen compensation,truly amazing!
rnb2r2/p3bpkp/1ppq3N/6p1/Q7/2P3P1/P4PBP/R1B1R1K1 b - - 0 19
The turning point that I found in the game was 32...Re8 in this position
rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 33
A truly remarkable game,but does your engine find or like 29.Qh3?

carldaman
Posts: 1738
Joined: Sat Jun 02, 2012 12:13 am

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by carldaman » Sun Dec 24, 2017 4:06 am

FWCC wrote:rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29
In this position will your engine find Qh3 which engines say is STILL EQUAL?
I used latest SF and the move was not a choice.Engines like 29Bd4+

This game is amazing to me due to this position:
rn3r2/p3b2p/2p3k1/1p3bp1/8/q1P1B1P1/P4PB1/3RR1KQ b - - 0 26
The Queen is positioned on H1 out of play.This looks like something from science fiction,an Alien game!After going over this game,the whole point was that Black's Queenside was NEVER developed,and here AlphaZero makes the TAL decision to hang the knight for unseen compensation,truly amazing!
rnb2r2/p3bpkp/1ppq3N/6p1/Q7/2P3P1/P4PBP/R1B1R1K1 b - - 0 19
The turning point that I found in the game was 32...Re8 in this position
rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 33
A truly remarkable game,but does your engine find or like 29.Qh3?
Just checked the first position, I'll look at the others later.

FEN: rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29

Zero-the-Hero:
1/1 00:00 158 158k +6.17 Qh1-e4 Qa3xa2
2/2 00:00 334 334k +14.32 Qh1-e4 Qa3xc3 Qe4xe7+ Kg7-g8 Be3xg5
3/4 00:00 527 527k +11.36 f2-f4 Qa3xc3 f4xg5
4/6 00:00 1k 1,302k +9.69 Be3-d4+ Be7-f6 Bd4xf6+ Kg7xf6 Qh1-e4 Kf6-g7
6/8 00:00 3k 1,473k +10.60 Be3-d4+ Rf8-f6 Re1-e6 Nb8-d7 Rd1-e1 Qa3xa2 Re6xe7+ Qa2-f7
7/10 00:00 8k 1,944k +8.35 f2-f4 Qa3xc3 f4xg5 Be7xg5 Be3-d4+ Qc3xd4+ Rd1xd4 c6-c5
8/12 00:00 20k 2,171k +6.27 Be3-d4+ Be7-f6 Re1-e3 g5-g4 Qh1-h4 Bf6xd4 c3xd4 Rf8-h8 Re3-e7+ Qa3xe7
20/37 00:00 2,474k 3,475k +0.85 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4 Rd2-e2 Qc5
-f5 Re6-e7+ Rf8-f7 Re7xf7+ Qf5xf7 Qh1-h4 Nb8-d7 Re2-e7 Nd7-e5 Kg1-g2 Kg7-g8 Re7xf7 Kg8xf7 Qh4-h7+ Kf7-e6

21/37 00:00 3,142k 3,415k +1.93 Qh1-h3 Be7-f6 Kg1-g2 Rf8-h8 Qh3-g4 Rh8-h5 Qg4-c8 Qa3xa2 Rd1-d6 Rh5-h8 Qc8-b7+ Qa2
-f7 Qb7xa8 Bf6-e5 Qa8xa7 Qf7xa7 Be3xa7 Be5xd6 Ba7-d4+ Kg7-g8 Re1-e8+ Bd6-f8 Bd4xh8 Nb8-d7 Re8-d8 Kg8xh8 Rd8xd7 Bf8-g7 Rd7
-c7 Bg7xc3 Rc7xc6 b5-b4 Rc6xg6

27/59+ 00:22 65,423k 2,967k +1.85 Qh1-h3
27/60 00:24 74,463k 2,984k +2.78 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4
Qh1-e4 Qc5-f5 Qe4-d4+ Kg7-g8 Re6-e7 Qf5-f6 Qd4-e3 Nb8-a6 Re7-e6 Qf6-f5 Rd2-d6 Kg8-g7 Qe3-d4+ Kg7-h6 Re6xg6+ Qf5xg6 Rd6xg6+
Kh6xg6 Qd4xg4+ Kg6-f6 Qg4-f3+ Kf6-g7 Qf3xc6 Ra8-d8 Qc6xa6 Rd8-d1+ Kg1-g2 Rd1-d2 Qa6xa7+ Rf8-f7 Qa7-c5 Rd2xf2+ Qc5xf2 Rf7xf2+ Kg2xf2 Kg7-f7 Kf2-f3 Kf7-f6 c3-c4 b5-b4 c4-c5 Kf6-e6 Kf3-e4 Ke6-d7

29/48- 00:38 117,287k 3,062k +1.72 Be3-d4+ Be7-f6
29/49+ 00:39 120,504k 3,074k +1.81 Qh1-h3
33/53- 01:35 322,085k 3,381k +4.23 Qh1-h3 Be7-f6

Cheers,
CL 8-)[/b]
Last edited by carldaman on Sun Dec 24, 2017 4:09 am, edited 1 time in total.

Geonerd
Posts: 66
Joined: Fri Mar 10, 2017 12:44 am

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by Geonerd » Sun Dec 24, 2017 4:08 am

Nope.

ASMFish, Oct 4, 2017. 24 core AMD Magny @ 2.3. Lg Pages. 16G hash. 34~35 Mn/s.
After 13+ minutes and 28G positions, the program rates Qe4, Bc1, Bd4+, and Qh3 equally, at +0.00

Maybe I'll let it think overnight. It makes a nice room heater! :wink:

Agree. Many of Monster Zero's games are quite impressive! :shock:

carldaman
Posts: 1738
Joined: Sat Jun 02, 2012 12:13 am

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by carldaman » Sun Dec 24, 2017 4:29 am

carldaman wrote:
FWCC wrote:rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29
In this position will your engine find Qh3 which engines say is STILL EQUAL?
I used latest SF and the move was not a choice.Engines like 29Bd4+

This game is amazing to me due to this position:
rn3r2/p3b2p/2p3k1/1p3bp1/8/q1P1B1P1/P4PB1/3RR1KQ b - - 0 26
The Queen is positioned on H1 out of play.This looks like something from science fiction,an Alien game!After going over this game,the whole point was that Black's Queenside was NEVER developed,and here AlphaZero makes the TAL decision to hang the knight for unseen compensation,truly amazing!
rnb2r2/p3bpkp/1ppq3N/6p1/Q7/2P3P1/P4PBP/R1B1R1K1 b - - 0 19
The turning point that I found in the game was 32...Re8 in this position
rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 33
A truly remarkable game,but does your engine find or like 29.Qh3?


AlphaZero - Stockfish 8, match 2017

"here AlphaZero makes the TAL decision to hang the knight for unseen compensation,truly amazing!"

[D]rnb2r2/p3bpkp/1ppq3N/6p1/Q7/2P3P1/P4PBP/R1B2RK1 w - - 0 1

Analysis by Zero-the-Hero:

19.Qe4 Kxh6 20.Re1 f6 21.Qxe7 Qxe7 22.Rxe7
+- (2.12) Depth: 7/8 00:00:00 2kN
19.Ng4 Rd8 20.Re1 Qc5 21.Re5 Qxc3 22.Rxe7 Qxa1
+- (2.13) Depth: 8/9 00:00:00 6kN
19.Qe4 Kxh6 20.Qh4+ Kg7 21.Bxg5 Bxg5 22.Qxg5+ Kh8 23.Rae1 Rd8 24.Qh4
+/= (0.62) Depth: 9/12 00:00:00 11kN
19.Re1 Kxh6 20.Qe4 Kg7 21.Qxe7 Qxe7 22.Rxe7 h6 23.Ba3 Bf5 24.Rd1 Be6 25.Bd6 Bxa2 26.Be5+ Kg8
+- (1.92) Depth: 10/16 00:00:00 22kN
19.Re1 Qxh6 20.Rxe7 Kg8 21.Qd4 Qg7 22.Qxg7+ Kxg7 23.Bxg5 Kg6 24.h4 Kg7 25.Rae1
+- (3.90) Depth: 11/21 00:00:00 55kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Qd4 Qxd4 22.cxd4 Be6 23.Bxg5 Bd5 24.Bf4 Bxg2 25.Kxg2 h6 26.Rae1
+- (3.11) Depth: 12/21 00:00:00 79kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Qd4 Qxd4 22.cxd4 Be6 23.Bxg5 Bd5 24.Rc1 Na6 25.Bxd5 cxd5 26.Bd2 Rfc8 27.Rxc8 Rxc8 28.Rxa7
+- (2.22) Depth: 13/21 00:00:00 114kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Qd4 Qxd4 22.cxd4 Be6 23.Bxg5 Bd5 24.Rc1 Bxg2 25.Kxg2 c5 26.dxc5 Nc6 27.Rd7 h6 28.Bf4
+- (2.51) Depth: 14/23 00:00:00 153kN
19.Re1 b5 20.Qe4 Bd8 21.Nf5+ Bxf5 22.Qxf5 Qg6 23.Qc5 Nd7 24.Qxc6 Qxc6 25.Bxc6 Ba5 26.Bxd7 Bxc3 27.Ba3 Bxa1 28.Bxf8+ Rxf8
+- (3.08) Depth: 15/24 00:00:00 575kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Bxc6 Nc5 23.Qd4 Bf5 24.Bxa8 Rxa8 25.Bxc5 bxc5 26.Qxc5 Rd8 27.Re5 Be6 28.Rxg5+ Kh8
+- (3.31) Depth: 16/24 00:00:00 898kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Bxc6 Nc5 23.Qd4 Be6 24.Bxa8 Rxa8 25.Qxf6+ Kxf6 26.Bxc5 bxc5 27.Rc7 h5 28.a4 a6 29.Rc6 h4 30.Kg2 hxg3 31.fxg3
+- (3.15) Depth: 17/29 00:00:00 1377kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Rae1 Nc5 23.Bxc5 bxc5 24.Bxc6 Rb8 25.Qxa7 Bf5 26.Bd5 Qxc3 27.Bxf7 Kh8 28.Bd5 Rb1
+- (2.97) Depth: 18/32 00:00:00 1982kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Rae1 Nc5 23.Bxc5 bxc5 24.Bxc6 Rb8 25.Qxa7 Bh3 26.Bd5 Kh8 27.Qxc5 Rbc8 28.Qd4 Qxd4 29.cxd4 Rcd8 30.Bxf7 Rxd4 31.Bh5 Rc8 32.Rf7
+- (2.70) Depth: 19/32 00:00:01 3169kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Qxc6 Bf5 23.Qxf6+ Kxf6 24.Bxa8 Rxa8 25.Rf1 Bh3 26.Rfe1 Be6 27.f4 h6 28.Bd6 Kg6 29.Kf2 Kh5 30.a3 Kg6 31.Kf3
+- (2.72) Depth: 20/38 00:00:01 4922kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Qxc6 Nc5 23.Bxc5 bxc5 24.Qxc5 Rb8 25.Rae1 Bf5 26.Qxa7 Qxc3 27.h4 Kh8 28.R7e3 Qf6 29.hxg5 Qxg5 30.Qc5 Qg4 31.Qc3+ f6
+- (2.68) Depth: 21/40 00:00:02 6780kN
19.Re1
+- (2.72 ++) Depth: 22/40 00:00:03 8956kN
19.Re1 Kxh6
+- (2.25 --) Depth: 22/40 00:00:03 9930kN
19.Re1 Kxh6
+- (2.25 --) Depth: 22/40 00:00:03 10031kN
19.Re1 Kxh6
+- (1.96 --) Depth: 22/40 00:00:03 10085kN
19.Re1
+- (2.21 ++) Depth: 22/40 00:00:03 10626kN
19.Re1
+- (2.71 ++) Depth: 22/42 00:00:04 11781kN
19.Re1 Qxh6 20.Rxe7 Qf6 21.Ba3 Na6 22.Qxc6 Nc5 23.Bxc5 bxc5 24.Qxc5 Rb8 25.Rae1 Bf5 26.Qxa7 Qxc3 27.h4 Kh8 28.R7e3 Qf6 29.hxg5 Qxg5 30.Re7 Qf6 31.a4
+- (2.66) Depth: 22/42 00:00:04 13313kN
19.Re1
+- (2.73 ++) Depth: 23/34 00:00:05 14522kN
19.Re1
+- (2.81 ++) Depth: 23/34 00:00:05 14583kN
19.Re1
+- (2.92 ++) Depth: 23/34 00:00:05 15882kN
19.Re1
+- (3.08 ++) Depth: 23/34 00:00:07 19969kN
19.Re1 Kxh6
+- (2.58 --) Depth: 23/37 00:00:07 20175kN
19.Re1 Kxh6
+- (2.29 --) Depth: 23/43 00:00:07 21280kN
19.Re1
+- (2.61 ++) Depth: 23/43 00:00:07 22812kN
19.Re1 Kxh6
+- (1.90 --) Depth: 23/46 00:00:08 23295kN
19.Re1 Kxh6 20.h4 Kg7 21.hxg5 Kg8 22.Qh4 Bf5 23.Bf4 Qa3 24.Rad1 Bc5 25.Be5 Nd7 26.Rxd7 Rae8 27.g6 Bxg6 28.Qf6 Rxe5 29.Qxe5 Qxa2 30.Qf6 Qc2 31.Rxa7 Qd3 32.Kh2 Qd2 33.Rf1 Qh6+ 34.Kg1 Qg7
+- (1.90) Depth: 23/46 00:00:08 23754kN
19.Re1
+- (1.98 ++) Depth: 24/41 00:00:08 25033kN
19.Re1
+- (2.05 ++) Depth: 24/41 00:00:08 25076kN
19.Re1
+- (2.17 ++) Depth: 24/41 00:00:08 25159kN
19.Re1
+- (2.32 ++) Depth: 24/41 00:00:08 25320kN
19.Re1 Kxh6
+- (1.83 --) Depth: 24/41 00:00:08 25901kN
19.Re1
+- (2.18 ++) Depth: 24/44 00:00:09 26321kN
19.Re1
+- (2.57 ++) Depth: 24/44 00:00:09 26782kN
19.Re1 Kxh6 20.h4 Kg7 21.hxg5 Kg8 22.Qh4 Bf5 23.Bf4 Qa3 24.Rad1 Bc5 25.Kh2 Nd7 26.Rxd7 Qxc3 27.Rd2 Rad8 28.Rde2 Bxf2 29.Rc1 Bxg3+ 30.Qxg3 Qxg3+ 31.Bxg3 f6 32.gxf6 Rxf6 33.Be5 Rg6 34.Rxc6 Rd3 35.Rxg6+ Bxg6 36.a3 Rxa3 37.Bd5+ Bf7
+- (1.96) Depth: 24/44 00:00:09 27210kN
19.Re1
+- (2.03 ++) Depth: 25/43 00:00:09 28702kN
19.Re1
+- (2.11 ++) Depth: 25/43 00:00:09 28922kN
19.Re1 Kxh6
+- (1.88 --) Depth: 25/43 00:00:10 29294kN
19.Re1
+- (2.05 ++) Depth: 25/43 00:00:10 29328kN
19.Re1 Kxh6 20.h4 Kg7 21.hxg5 Kg8 22.Qh4 Bf5 23.Bf4 Qa3 24.Rad1 Bc5 25.Be5 Nd7 26.Rxd7 Rae8 27.g6 Bxg6 28.Qf6 Rxe5 29.Qxe5 Qxa2 30.Qf6 Qc2 31.Rxa7 Qd3 32.Kh2 Qd2 33.Rf1 Qh6+ 34.Kg1 Qg7 35.Qf4 Kh8 36.Bxc6 Qxc3 37.Rc1 Qd4 38.Qxd4+ Bxd4 39.Rd1 Rc8 40.Rxd4
+- (2.07) Depth: 25/43 00:00:10 29648kN
19.Re1
+- (2.14 ++) Depth: 26/49 00:00:10 31683kN
19.Re1
+- (2.22 ++) Depth: 26/49 00:00:11 32063kN
19.Re1 Kxh6 20.h4 Kg7 21.hxg5 Kg8 22.Qh4 Bf5 23.Bf4 Qa3 24.Rad1 Bc5 25.Be5 Nd7 26.Rxd7 Rae8 27.g6 Bxg6 28.Qf6 Rxe5 29.Qxe5 Qxa2 30.Qf6 Qc2 31.Rxa7 Qd3 32.Kh2 Qd2 33.Rf1 Qh6+ 34.Kg1 Qg7 35.Qxg7+ Kxg7 36.Bxc6 Rd8 37.Bd7 Be7 38.Kg2 Bf6 39.Rc1 Kh8 40.c4 Kg7 41.Rh1 Bd3
+- (2.07) Depth: 26/49 00:00:11 32872kN
19.Re1
+- (2.14 ++) Depth: 27/49 00:00:11 34090kN
19.Re1
+- (2.22 ++) Depth: 27/49 00:00:12 35253kN
19.Re1
+- (2.33 ++) Depth: 27/49 00:00:12 35757kN
19.Re1
+- (2.49 ++) Depth: 27/49 00:00:12 37091kN
19.Re1
+- (2.70 ++) Depth: 27/49 00:00:13 38446kN
19.Re1 Kxh6
+- (1.99 --) Depth: 27/49 00:00:14 43922kN
19.Re1 Kxh6
+- (1.61 --) Depth: 27/50 00:00:17 51399kN
19.Re1 Kxh6
+/- (1.11 --) Depth: 27/60 00:00:23 70591kN
19.Re1 Kxh6 20.h4 f6 21.hxg5+ fxg5 22.Qh4+ Kg6 23.Be4+ Bf5 24.Bxf5+ Rxf5 25.Qe4 Qd5 26.Qxe7 Nd7 27.Bxg5 Rf7 28.Re6+ Kg7 29.Bh6+ Kg8 30.Qg5+ Qxg5 31.Bxg5 Raf8 32.Re2 Nc5 33.Rd1 Rd7 34.Rxd7 Nxd7 35.Re7 Rf7 36.Re8+ Rf8 37.Re6 c5 38.f4 c4 39.Re7 Rf7 40.Kf2 Rxe7 41.Bxe7
+/- (0.89) Depth: 27/60 00:00:28 86597kN
19.Re1
+/- (0.97 ++) Depth: 28/53 00:00:30 94317kN
19.Re1
+/- (1.04 ++) Depth: 28/53 00:00:30 95103kN
19.Re1 Kxh6 20.h4 f6 21.hxg5+ fxg5 22.Qh4+ Kg6 23.Qe4+ Bf5 24.Qxe7 Qxe7 25.Rxe7 Na6 26.Bxc6 Rac8 27.Rxa7 Rxc6 28.Rxa6 Bd3 29.Ra7 Rxc3 30.Be3 Rf7 31.Ra8 Rc6 32.a4 Re6 33.Rg8+ Rg7 34.Rxg7+ Kxg7 35.Bxg5 Bc2 36.Be3 h6 37.g4 Bd3
+/- (0.99) Depth: 28/53 00:00:31 97909kN
19.Re1
+/- (1.07 ++) Depth: 29/48 00:00:34 106mN
19.Re1
+/- (1.14 ++) Depth: 29/48 00:00:34 107mN
19.Re1
+/- (1.25 ++) Depth: 29/48 00:00:35 109mN
19.Re1
+/- (1.41 ++) Depth: 29/48 00:00:35 111mN
19.Re1 Kxh6 20.h4 f6 21.Be3 Kg7 22.Rad1 Qe6 23.hxg5 Qf7 24.Bd4 Kg8 25.Rxe7 Qxe7 26.Bxf6 Rxf6 27.gxf6 Qxf6 28.Qe4 Nd7 29.Qxc6 Qxc6 30.Bxc6 Rb8 31.Bxd7 Kf8 32.Rd4 a5 33.Bxc8 Rxc8 34.c4 Rc7 35.Rh4 Kg7 36.Kg2 Rf7 37.f4
+/- (1.49) Depth: 29/48 00:00:37 117mN
19.Re1
+/- (1.57 ++) Depth: 30/44 00:00:39 124mN
19.Re1
+- (1.64 ++) Depth: 30/44 00:00:40 126mN
19.Re1
+- (1.75 ++) Depth: 30/44 00:00:41 128mN
19.Re1 Kxh6 20.h4 f6 21.Be3 Kg7 22.Rad1 Qe6 23.hxg5 Qf7 24.Bf4 fxg5 25.Rxe7 Qxe7 26.Qd4+ Kg8 27.Bd6 Qf6 28.Bxf8 Qxd4 29.Rxd4 Kxf8 30.Rd8+ Ke7 31.Rxc8 Kd7 32.Rh8 Kd6 33.Rxh7 Nd7 34.Rh6+ Ke7 35.Bxc6 Rd8 36.Rh7+ Kd6 37.Ba4 Nc5 38.Rxa7 Ne4 39.Bb3 Nxc3 40.Kg2 Rf8
+- (1.65) Depth: 30/50 00:00:45 140mN
19.Re1
+- (1.73 ++) Depth: 31/43 00:00:47 149mN

carldaman
Posts: 1738
Joined: Sat Jun 02, 2012 12:13 am

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by carldaman » Sun Dec 24, 2017 4:31 am

Reposted with diagram included:

[D]rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29

Zero-the-Hero:
1/1 00:00 158 158k +6.17 Qh1-e4 Qa3xa2
2/2 00:00 334 334k +14.32 Qh1-e4 Qa3xc3 Qe4xe7+ Kg7-g8 Be3xg5
3/4 00:00 527 527k +11.36 f2-f4 Qa3xc3 f4xg5
4/6 00:00 1k 1,302k +9.69 Be3-d4+ Be7-f6 Bd4xf6+ Kg7xf6 Qh1-e4 Kf6-g7
6/8 00:00 3k 1,473k +10.60 Be3-d4+ Rf8-f6 Re1-e6 Nb8-d7 Rd1-e1 Qa3xa2 Re6xe7+ Qa2-f7
7/10 00:00 8k 1,944k +8.35 f2-f4 Qa3xc3 f4xg5 Be7xg5 Be3-d4+ Qc3xd4+ Rd1xd4 c6-c5
8/12 00:00 20k 2,171k +6.27 Be3-d4+ Be7-f6 Re1-e3 g5-g4 Qh1-h4 Bf6xd4 c3xd4 Rf8-h8 Re3-e7+ Qa3xe7
20/37 00:00 2,474k 3,475k +0.85 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4 Rd2-e2 Qc5
-f5 Re6-e7+ Rf8-f7 Re7xf7+ Qf5xf7 Qh1-h4 Nb8-d7 Re2-e7 Nd7-e5 Kg1-g2 Kg7-g8 Re7xf7 Kg8xf7 Qh4-h7+ Kf7-e6

21/37 00:00 3,142k 3,415k +1.93 Qh1-h3 Be7-f6 Kg1-g2 Rf8-h8 Qh3-g4 Rh8-h5 Qg4-c8 Qa3xa2 Rd1-d6 Rh5-h8 Qc8-b7+ Qa2
-f7 Qb7xa8 Bf6-e5 Qa8xa7 Qf7xa7 Be3xa7 Be5xd6 Ba7-d4+ Kg7-g8 Re1-e8+ Bd6-f8 Bd4xh8 Nb8-d7 Re8-d8 Kg8xh8 Rd8xd7 Bf8-g7 Rd7
-c7 Bg7xc3 Rc7xc6 b5-b4 Rc6xg6

27/59+ 00:22 65,423k 2,967k +1.85 Qh1-h3
27/60 00:24 74,463k 2,984k +2.78 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4
Qh1-e4 Qc5-f5 Qe4-d4+ Kg7-g8 Re6-e7 Qf5-f6 Qd4-e3 Nb8-a6 Re7-e6 Qf6-f5 Rd2-d6 Kg8-g7 Qe3-d4+ Kg7-h6 Re6xg6+ Qf5xg6 Rd6xg6+
Kh6xg6 Qd4xg4+ Kg6-f6 Qg4-f3+ Kf6-g7 Qf3xc6 Ra8-d8 Qc6xa6 Rd8-d1+ Kg1-g2 Rd1-d2 Qa6xa7+ Rf8-f7 Qa7-c5 Rd2xf2+ Qc5xf2 Rf7xf2+ Kg2xf2 Kg7-f7 Kf2-f3 Kf7-f6 c3-c4 b5-b4 c4-c5 Kf6-e6 Kf3-e4 Ke6-d7

29/48- 00:38 117,287k 3,062k +1.72 Be3-d4+ Be7-f6
29/49+ 00:39 120,504k 3,074k +1.81 Qh1-h3
33/53- 01:35 322,085k 3,381k +4.23 Qh1-h3 Be7-f6

carldaman
Posts: 1738
Joined: Sat Jun 02, 2012 12:13 am

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by carldaman » Sun Dec 24, 2017 4:47 am

FWCC wrote: The turning point that I found in the game was 32...Re8 in this position
rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 33
Zero thinks Sf was busted before Re8 was played. At that point several moves were probably winning:

AlphaZero - Stockfish 8, 1 min / 1 move
[D]rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 1

Analysis by Zero-the-Hero: (Depth 30)

1. +- (2.81): 33.Rd6 Be5 34.Rd3 Bf6 35.cxb5 cxb5 36.Qh6+ Kf7 37.Bxg5 Qh8 38.Rd6 Qxh6 39.Rxf6+ Kg7 40.Rxh6 Nd7 41.Rfxg6+ Kf7 42.Rd6 Nf8 43.Rd5 Rab8 44.Rf5+ Kg7 45.Rh4 Rb6 46.Bc1 Ng6 47.Bb2+ Kg8 48.Rg5 Kf8 49.Rhg4 Ree6 50.Bd4 Rbc6 51.Rxb5 Ke8 52.Rb8+ Kd7 53.Be3 a5 54.Rb5 Ke8 55.Rxa5 Rcd6 56.Rb4 Ne5 57.Rb7 Nc6 58.Ra8+ Rd8
2. +- (2.61): 33.Rd2 Re4 34.Rd6 Be7 35.Rd4 Re6 36.Rd2 bxc4 37.Bd4+ Kf7 38.Qg4 Nd7 39.Rh8 Qxh8 40.Bxh8 Nf6 41.Bxf6 Bxf6 42.Qxc4 Rd8 43.Rc2 a5 44.Qb3 Rd5 45.Rc1 g4 46.Rxc6 Rxc6 47.Qxd5+ Re6 48.Qxa5 Re5 49.Qc7+ Kg8 50.Qc8+ Kh7 51.Qxg4 Bg7 52.Qc4 Rh5 53.Qc6 Rf5 54.Qc7 Kh6 55.Qd7
3. +- (2.12): 33.Rd3 Re4 34.Rd6 Be7 35.Rd4 Re6 36.Rdd1 Bf6 37.Rd2 a5 38.Bd4 Kf7 39.Bxf6 Rxf6 40.Re2 Ra7 41.Qh6 Na6 42.Qh7+ Qxh7 43.Rxh7+ Kf8 44.Rxa7 Nb4 45.Rd2 Rf7 46.Ra8+ Ke7 47.cxb5 cxb5 48.Rxa5 Rf5 49.g4 Rc5 50.Rb2 Rc4 51.f3 Nd5 52.Raxb5 Ne3+ 53.Kh2
4. +- (2.07): 33.Bd4 Bxd4 34.Rxd4 Rd8 35.Qh6+ Kf7 36.Re4 Re8 37.Rxe8 Qxe8 38.Rd1 Qf8 39.Qh7+ Qg7 40.Qh3 Qf8 41.Re1 Kg7 42.Qe6 Qf6 43.Qc8 bxc4 44.Qb7+ Nd7 45.Qxa8 Ne5 46.Qxa7+ Nf7 47.Qc5 c3 48.Re8 Kh6 49.Re3 g4 50.Qxc3 Qxc3 51.Rxc3 Ne5 52.Rc5 Nd3 53.Rxc6 Kg5 54.Rb6 Kf5
5. +- (2.18): 33.Qh6+ Kf7 34.Bxg5 Qg7 35.Qh4 Re5 36.Bxf6 Qxf6 37.Qh7+ Qg7 38.Qh3 Na6 39.Rd4 Nc5 40.Rh4 Rae8 41.cxb5 cxb5 42.Rh7 Rh5 43.Rxg7+ Kxg7 44.Qg4 Rxh1 45.Kxh1 Ne6 46.Qe4 Re7 47.Qe5+ Kf7 48.Qxb5 Nc7 49.Qc4+ Ne6 50.Kg2 Rc7 51.Qb5 Kf6 52.f4 Rc5 53.Qa6 Rc7 54.Kh3 Kf7 55.Kg2 Rc2+ 56.Kh3
6. +/- (1.57): 33.Bc1 Re6 34.Rh2 Re4 35.Rd6 Kf7 36.cxb5 Be7 37.Rd3 g4 38.Qh6 Na6 39.bxa6 Qg7 40.Qxg7+ Kxg7 41.Rd7 Kf6 42.Rc7 c5 43.Rh7 Bd6 44.Bb2+ Ke6 45.Rcg7 Kd5 46.Rxg6 Re2 47.Bc1 Rf8 48.Be3 Re8 49.Rd7 R8xe3 50.Rgxd6+ Kc4 51.Rxa7 Ra2 52.Rg6 Re1

peter
Posts: 1786
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by peter » Sun Dec 24, 2017 8:12 am

Hi!
FWCC wrote:rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29
All you have to do is to raise komodo's Dynamism to 200 (instead of 130), everything else can be left by Default except switching off tbs:

[d]rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 1
Analysis by Komodo 11.2.2 64-bit:

29.Bd4+ Bf6
= (0.25 --) Depth: 7 00:00:00 618kN
...
29.Bc1 Qc5 30.Be3 Qa3
= (0.00) Depth: 27 00:00:01 20273kN
29.Bc1 Qc5 30.Be3 Qa3
= (0.00) Depth: 28 00:00:01 26640kN
29.Qh3 Bf6 30.Kg2
= (0.06 ++) Depth: 28 00:00:04 66857kN
...
29.Qh3 Bf6 30.Kg2
+/- (1.30 ++) Depth: 28 00:00:23 376MN
29.Qh3 Bf6 30.Kg2 Qxa2 31.Rh1 Qg8 32.c4 Re8 33.Bd4 Bxd4 34.Rxd4 Rd8 35.Rxd8 Qxd8 36.Qe6 Nd7 37.Rd1 Nf8 38.Rxd8 Nxe6 39.Rxa8 b4 40.Rxa7+ Kf6 41.Kf3 b3 42.Rb7 Ke5 43.Ke3 Nc5 44.Rb6 g4 45.Rb4 Ke6 46.Kd4 Kd6 47.Rb6 Ne6+ 48.Kc3 Nc5 49.Rb8
+/- (1.27) Depth: 28 00:00:24 395MN
...
29.Qh3 Bf6 30.Kg2 Qxa2 31.Rh1 Qg8 32.c4 Re8 33.Bd4 Bxd4 34.Rxd4 Rd8 35.Rxd8 Qxd8 36.Qe6 Nd7 37.Rd1 Nc5 38.Rxd8 Nxe6 39.Rxa8 Kf6 40.Rxa7 Ke5 41.cxb5 cxb5 42.Rb7 Nd4 43.Kh3 Kf5 44.f3 Nxf3 45.Rxb5+ Ne5 46.Kg2 Ke4 47.Kf2 Kd4 48.Rb4+ Kd3 49.Ra4 Nc4 50.g4 Ne5 51.Kg3 Nd7 52.Rb4 Ne5 53.Rb5
+/- (1.26) Depth: 33 00:02:12 2223MN
...
29.Qh3 Bf6 30.Kg2 Rh8 31.Qg4 Rf8 32.Bd4 Qd6 33.Be5 Qxe5 34.Rxe5 Bxe5 35.f4 gxf4 36.gxf4 Bf6 37.f5 g5 38.Rd6 Rf7 39.Qh5 Nd7 40.Qg6+ Kf8 41.Rxd7 Rxd7 42.Qxf6+ Kg8 43.Qxc6 Rad8 44.Kg3 Rd5 45.Qe6+ Kf8 46.Qf6+ Ke8 47.Qxg5 R8d7 48.Qg8+ Ke7 49.Kf4 a5 50.Qe6+ Kf8 51.Qc6 Kg7 52.Ke4 Rd2 53.Qxb5 Re7+ 54.Kf3 Rxa2 55.Qc6 Ree2 56.Qg6+ Kf8 57.Qf6+ Kg8 58.Qd8+ Kf7 59.Qd5+
+/- (1.32) Depth: 34 00:03:59 4091MN
Tried 3 times, always at such a few seconds only the move came up.

It's a really nice one and can be taken as a test position for engines and their settinigs, SugaR needs more Midgame King's Safety only e.g., poor Houdini needs both, Tactical 1 and Contempt raised to get it really quickly, yet with Contempt 10 and Tactical 1 it's within few seconds only too.

Here is latest SF with Contempt 50.
(100 is much faster, especially as for getting c4 later on into output- line as for 30...Qxa2 (?!) or f4 as for 30...Rh8 together with the winning eval.)

Contempt 50 just to show that no settings are necessary, that cannot be taken for game playing still:
rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 1

Analysis by Brainfish 191217 64 POPCNT:

29.Bc1 Qc5 30.Be3 Qa3 31.Bc1
= (0.00) Depth: 7/6 00:00:05 25kN
...
29.Bc1 Qc5 30.Be3 Qa3 31.Bc1
= (0.00) Depth: 37/6 00:00:13 196MN
29.Qh3
= (0.07 ++) Depth: 38/53 00:00:25 506MN
...
29.Qh3 Bf6 30.Kg2 Qxa2 31.Rh1 Qg8 32.Rd6 Re8 33.Qh6+ Kf7 34.Bxg5 Be5 35.Rd3 Qg7 36.Qh4 Kg8 37.f4 Bxc3 38.Rxc3 Re2+ 39.Kf1 Re1+ 40.Kxe1 Qxc3+ 41.Kf2 Qc2+ 42.Ke3 Qc3+ 43.Kf2
= (0.00) Depth: 38/72 00:01:12 1674MN
...
29.Qh3 Bf6 30.Kg2 Rh8 31.Qg4 Rf8 32.Bd4 Qd6 33.Be5 Qxe5 34.Rxe5 Bxe5 35.f4 gxf4 36.gxf4 Bf6 37.f5 g5 38.Qh5 Rf7 39.Qg6+ Kf8 40.Rd6 Nd7 41.Rxd7 Rxd7 42.Qxf6+ Kg8 43.Qxc6 Rad8 44.Kg3 Rf7 45.Qxb5 Rd1 46.Qe8+ Kg7 47.Qe5+ Kf8 48.Kg4 Rf1 49.Qc5+ Kg8 50.Qc8+ Kg7 51.Kxg5 Rg1+ 52.Kf4 Rf1+ 53.Ke4 Re1+ 54.Kd4 Rd1+ 55.Kc4 Rf1 56.Kb4 R7xf5 57.Qc7+ Rf7 58.Qe5+ Kg6 59.Qe8 Kg5 60.Qe3+ Kf6 61.c4 Rb7+ 62.Kc3
+/- (1.30) Depth: 39/86 00:05:41 8493MN
29.Qh3 Bf6
...
29.Bc1
+/- (1.30 ++) Depth: 40/76 00:06:20 9495MN
...
29.Bc1 Qc5 30.Be3 Qa3 31.Qh3 Bf6 32.Kg2 Rh8 33.Qg4 Rf8 34.Bd4 Qd6 35.Be5 Qxe5 36.Rxe5 Bxe5 37.f4 gxf4 38.gxf4 Bf6 39.f5 g5 40.Qh5 Rf7 41.Qg6+ Kf8 42.Rd6 Nd7 43.Rxd7 Rxd7 44.Qxf6+ Kg8 45.Qxc6 Rad8 46.Kg3 Rd5 47.Kg4 R8d7 48.Kxg5 Kf7 49.Kf4 R7d6 50.Qb7+ Rd7 51.Qa6 Ke8 52.Qg6+ Kf8 53.Qc6 Kf7 54.Ke4 R5d6 55.Qxb5 Rc7 56.Ke5 Rdc6 57.Kf4 Rxc3 58.Qd5+ Kg7 59.Kg5 R3c6 60.Qe4 Kf7 61.Qh1 Kg7 62.Qf3 Rc5 63.Qe4 R5c6 64.Qe8
+/- (1.20) Depth: 40/76 00:06:51 10296MN
...
29.Bc1 Qc5 30.Be3 Qa3 31.Qh3 Bf6 32.Kg2 Rh8 33.Qg4 Rf8 34.Bd4 Qd6 35.Be5 Qxe5 36.Rxe5 Bxe5 37.f4 gxf4 38.gxf4 Bf6 39.f5 g5 40.Qh5 Rf7 41.Qg6+ Kf8 42.Rd6 Nd7 43.Rxd7 Rxd7 44.Qxf6+ Kg8 45.Qxc6 Rad8 46.Kg3 Rd5 47.Kg4 R8d7 48.Kxg5 Kf8 49.Kf4 Kf7 50.Ke4 R5d6 51.Qxb5 Rc7 52.Qb3+ Kg7 53.Kf4 Rdc6 54.Kg5 Rxc3 55.Qe6 Rg3+ 56.Kf4 Rgc3 57.Qg6+ Kh8 58.Qf6+ Kg8 59.Qd8+ Kg7 60.Kg5 Rg3+ 61.Kh4 Rgc3 62.Qg5+ Kf8 63.Qf6+ Kg8 64.Qg6+ Kf8 65.Qd6+ Kg8 66.Qd5+ Kg7 67.Kg5 Rg3+ 68.Kh5 Rh3+ 69.Kg4 Rhc3 70.Qe5+ Kh7 71.Qe6
+/- (1.09) Depth: 43/87 00:08:33 12897MN
29.Qh3
+/- (1.16 ++) Depth: 44/84 00:09:43 14687MN
...
29.Qh3 Bf6 30.Kg2 Qxa2 31.Rh1 Qg8 32.Rd6 Re8 33.c4 Be5 34.Bd4 Bxd4 35.Rxd4 Rd8 36.Rxd8 Qxd8 37.Qe6 Nd7 38.Rd1 Nc5 39.Rxd8 Nxe6 40.Rxa8 bxc4 41.Rxa7+ Kf6 42.Ra4 c3 43.Rc4 c5 44.Kf3 Ke5 45.Rxc3 Kd5 46.Rd3+ Nd4+ 47.Ke3 Ke6 48.Rc3 Kd5 49.Kd2 Nf5 50.Ra3 Nd6 51.Ra6 Ne4+ 52.Ke3 Nd6 53.Kf3 Ke5 54.Ra2 c4 55.Ke3
+/- (1.49) Depth: 44/84 00:13:32 20487MN

BTW Qh3 can be delayed by Bc1 , so the solution isn't a single best move one neither.


And to keep my feet on the ground as for "extraterrestric" moves, which these 4 or five of them are often called to be read in fora and seen in YouTubes again and again, which all deal with these same few moves, no wonder, we still haven't seen more than these 10 games still.

As I wrote somewhere else already:

If such a move arises in a corr.- game, any critical corr.-player would say, if Black wouldn't have played 26...Bg6 (?) the move wouldn't have been possible and further on Black didn't find the best answers neither. 30.Qxa2 was ?! too, even if the position might have been lost already anyhow,
Peter.

User avatar
mclane
Posts: 16277
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by mclane » Sun Dec 24, 2017 12:01 pm

The clue is to decide when to sac material for getting idealistic compensation.
AZ plays chess as if it has a fluent movement of material and spirit.

For stockfish it looks very very statically to me.

To quote a famous man:
I call this type of search Artificial Stupidity (AS). Since all the current programs operate in this way, ELO grading lists and inter-program tournaments are no more than a reflection of the partially-sighted playing the blind, whose AS algorithm is most efficient, but it is not
chess.
The classical program play chess as if it were the First World War in the trenches, no concept of mobility, no concept of cooperation of forces, no concept of knocking the enemy off balance with well timed blows; just material and pawn structure - if it plays boring chess, that's why - if it blunders against club players, that's why. It understands nothing of consequence.

Complete Chess System 2 - TAL
=============================

Classical paradigm
==================
When should we expect a major breakthrough in science ?
When will a lone developer 'step through the looking-glass' ?
Who will this developer be ?

The answer to the above two questions is of course whenever the old, classical programmers say 'we've reached perfection, there is no way to improve'; when the old paradigm says 'there is only one way'; when all the developers produce roughly equal results.

This is the situation we have today with chess programs. The classical paradigm is represented by Fritz3: fast and simple evaluation, pre-processing of the position before the search; and all strength, all hopes, in the search - nodes per second and search efficiency are the buzzwords.

For a classical program, to keep the search fast, the evaluation at each node must, of necessity, be brief. This evaluation is usually no more than a weighting given for each piece on each square (for example a knight might be worth 3.3 pawns on centre squares and 2.9 pawns on edge squares) and evaluation of the pawn structure for doubled pawns, passed pawns etc..
The classical pre-processing function looks for themes in the position and adjusts the square weightings accordingly - for example, if a knight is attacking a square next to the king, then increase the weighting for all the squares that the queen could cooperate with the knight in making a king attack, increase the knight weighting to keep it on the original square, increase other cooperating piece weightings and so on.

There is no doubt that this approach works but it cannot be the way forward. Preprocess- ing knowledge becomes more stupid with increasing search depth, as positions deep in the search tree becomes more removed from the assumptions of the original position,
the square weighting adjustments become more irrelevant (why weight the squares for the queen after the cooperating knight has been removed from the board ?- but the classical paradigm doesn't understand that !).

I call this type of search Artificial Stupidity (AS). Since all the current programs operate in this way, ELO grading lists and inter-program tournaments are no more than a reflection of the partially-sighted playing the blind, whose AS algorithm is most efficient, but it is not
chess.

They don't even know that they don't know
=========================================
Classic programs have static knowledge only, dynamic knowledge is beyond
the fast and simple evaluation function.

Statics:
- Material
- Structure
- Chronic weaknesses
- and more

Dynamics:
- Lead in development
- More active piece placement
- A specific and cooperative concentration of pieces
in a certain sector of the board.
- and more

Static features tend to be stable, they remain with time. Dynamic features can be dissipated with time. Static features are easy to calculate, classical programs include them. Dynamic features are difficult to calculate, they rely on interaction between the pieces, 'looking-glass' programs will begin to include them. And it is the lack
of the difficult dynamic feature calculation that marks the classical programs with so many bad games and bad moves - the types of games that allow GM's to laugh at chess pro- grams.

As GM John Nunn says 'the top programs occasionally win games against grandmasters, but they habitually lose games against ordinary club players, often making the most appalling anti-positional moves in the process.' What else does can he expect ? The old classical program finds a 24 move deep check thread, gets to the end of the thread, finds
it is not yet mate, and all it can do is add up the material, evaluate the pawn structure and return a score that shows absolutely no concept of the position !

To play chess without knowledge of chess is not to play chess, strong players will always beat such programs with superior knowledge.
The classical program play chess as if it were the First World War in the trenches, no concept of mobility, no concept of cooperation of forces, no concept of knocking the enemy off balance with well timed blows; just material and pawn structure - if it plays boring chess, that's why - if it blunders against club players, that's why. It understands nothing of consequence.

The philosophers of classical search claim that search finds everything and knows everything - they give as an example the knight fork: Without search the program knows that it is good to capture the queen with the knight. With three ply the search knows that it is good to knight fork the king and the queen. With five ply the search knows it is good to
play the knight to a position where it can threaten a fork and so on.
But the point must surely be that the search only has this knowledge within the tree. At the leaf nodes it has no such knowledge. An intelligent program can calculate as part of its evaluation function whether a knight fork is available; thus the intelligent program has
this knowledge distributed evenly over the entire search tree. In this way intelligence can replace search.
It is important here to distinguish between combinational knowledge and dynamic knowledge. In our example of the knight fork above, the classical program only has this 'knowledge' if the situation arises in tactics - the classical program only generates this knowledge as part of a combination to win the queen. If this win of the queen does not emerge from the search, then the knowledge does not exist !
The situation is perhaps clearer (and more serious) in the case of a king attack. If the classical program can find mate or win of material by some line attacking the king, in such case it has knowledge of the king attack; but if, at the search horizon, it has a strong attack, but not yet any material won, or king mated, it does not know this is a good line !
The 'looking-glass' program can calculate the attack strength FROM ITS EVALUATION FUNCTION. So, without actually finding mate or material win, the looking-glass program has the dynamic knowledge of the attack.
The classical program has combinational knowledge only by resolution of material within the search horizon. The looking-glass program has dynamic knowledge from its evaluation function. The looking-glass program is a planner, the classical program is a finder. The looking-glass program is pro-active, it makes plans to exploit the position; the classical program is re-active, it waits for a mistake by its opponent and then exploits it.
Who will be the developer ?
===========================
To answer our third question - 'who will be the developer ?', it is necessary to look at the personality of the classical programmers and their hangers-on. These programmers are characterised by a failure to show their emotions (do they ever smile), fear (just watch them operating at tournaments), refusal to discuss how their programs work (just try talking to them) , aversion to taking risks. It has always surprised me that the 'top' programmers are not good chess players. The hangers-on only make a little money, they jealously support their chosen proteges, and viciously attack their opponents. The hangers-on know little, pretend to know much and are governed by fear and greed.
Overall the impression is of a static, non-risk taking, hostile, World War I environment. The new paradigm will come from an unexpected quarter. From a developer with extrovert personality, accustomed to taking risks, a developer with chess knowledge, probably someone unpopular with the classical paradigm supporters, certainly unpopular
with the hangers-on and computer chess entourage. This developer will have been and certainly will be furiously attacked by the classicists.

Complete Chess System 2 - TAL
=============================

Classical paradigm
==================
When should we expect a major breakthrough in science ?
When will a lone developer 'step through the looking-glass' ?
Who will this developer be ?

The answer to the above two questions is of course whenever the old, classical programmers say 'we've reached perfection, there is no way to improve'; when the old paradigm says 'there is only one way'; when all the developers produce roughly equal results.

This is the situation we have today with chess programs. The classical paradigm is represented by Fritz3: fast and simple evaluation, pre-processing of the position before the search; and all strength, all hopes, in the search - nodes per second and search efficiency are the buzzwords.

For a classical program, to keep the search fast, the evaluation at each node must, of necessity, be brief. This evaluation is usually no more than a weighting given for each piece on each square (for example a knight might be worth 3.3 pawns on centre squares and 2.9 pawns on edge squares) and evaluation of the pawn structure for doubled pawns, passed pawns etc..
The classical pre-processing function looks for themes in the position and adjusts the square weightings accordingly - for example, if a knight is attacking a square next to the king, then increase the weighting for all the squares that the queen could cooperate with the knight in making a king attack, increase the knight weighting to keep it on the original square, increase other cooperating piece weightings and so on.

There is no doubt that this approach works but it cannot be the way forward. Preprocess- ing knowledge becomes more stupid with increasing search depth, as positions deep in the search tree becomes more removed from the assumptions of the original position,
the square weighting adjustments become more irrelevant (why weight the squares for the queen after the cooperating knight has been removed from the board ?- but the classical paradigm doesn't understand that !).

I call this type of search Artificial Stupidity (AS). Since all the current programs operate in this way, ELO grading lists and inter-program tournaments are no more than a reflection of the partially-sighted playing the blind, whose AS algorithm is most efficient, but it is not
chess.

They don't even know that they don't know
=========================================
Classic programs have static knowledge only, dynamic knowledge is beyond
the fast and simple evaluation function.

Statics:
- Material
- Structure
- Chronic weaknesses
- and more

Dynamics:
- Lead in development
- More active piece placement
- A specific and cooperative concentration of pieces
in a certain sector of the board.
- and more

Static features tend to be stable, they remain with time. Dynamic features can be dissipated with time. Static features are easy to calculate, classical programs include them. Dynamic features are difficult to calculate, they rely on interaction between the pieces, 'looking-glass' programs will begin to include them. And it is the lack
of the difficult dynamic feature calculation that marks the classical programs with so many bad games and bad moves - the types of games that allow GM's to laugh at chess pro- grams.

As GM John Nunn says 'the top programs occasionally win games against grandmasters, but they habitually lose games against ordinary club players, often making the most appalling anti-positional moves in the process.' What else does can he expect ? The old classical program finds a 24 move deep check thread, gets to the end of the thread, finds
it is not yet mate, and all it can do is add up the material, evaluate the pawn structure and return a score that shows absolutely no concept of the position !

To play chess without knowledge of chess is not to play chess, strong players will always beat such programs with superior knowledge.
The classical program play chess as if it were the First World War in the trenches, no concept of mobility, no concept of cooperation of forces, no concept of knocking the enemy off balance with well timed blows; just material and pawn structure - if it plays boring chess, that's why - if it blunders against club players, that's why. It understands nothing of consequence.

The philosophers of classical search claim that search finds everything and knows everything - they give as an example the knight fork: Without search the program knows that it is good to capture the queen with the knight. With three ply the search knows that it is good to knight fork the king and the queen. With five ply the search knows it is good to
play the knight to a position where it can threaten a fork and so on.
But the point must surely be that the search only has this knowledge within the tree. At the leaf nodes it has no such knowledge. An intelligent program can calculate as part of its evaluation function whether a knight fork is available; thus the intelligent program has
this knowledge distributed evenly over the entire search tree. In this way intelligence can replace search.
It is important here to distinguish between combinational knowledge and dynamic knowledge. In our example of the knight fork above, the classical program only has this 'knowledge' if the situation arises in tactics - the classical program only generates this knowledge as part of a combination to win the queen. If this win of the queen does not emerge from the search, then the knowledge does not exist !
The situation is perhaps clearer (and more serious) in the case of a king attack. If the classical program can find mate or win of material by some line attacking the king, in such case it has knowledge of the king attack; but if, at the search horizon, it has a strong attack, but not yet any material won, or king mated, it does not know this is a good line !
The 'looking-glass' program can calculate the attack strength FROM ITS EVALUATION FUNCTION. So, without actually finding mate or material win, the looking-glass program has the dynamic knowledge of the attack.
The classical program has combinational knowledge only by resolution of material within the search horizon. The looking-glass program has dynamic knowledge from its evaluation function. The looking-glass program is a planner, the classical program is a finder. The looking-glass program is pro-active, it makes plans to exploit the position; the classical program is re-active, it waits for a mistake by its opponent and then exploits it.

Dynamic knowledge v. Combinational knowledge
============================================

Oxford Softworks CCS2-v9.0
White: CCS2 486/33
Black: Genius2 486/33
Venue: 1 minute per move
Comment: 1-0

1. e4 e6
2. d4 d5 1
3. Nc3 Nf6 3
4. Bg5 Be7 5
5. e5 Nfd7 8
6. h4 Bxg5
7. hxg5 { CCS2's opening book ends }
.... Qxg5
8. Nf3 Qd8 { Genius2's opening book ends }
9. Bd3 h6
10. Qd2 { CCS2's dynamic knowledge - preventing O-O because
of the threat of Rxh6 }
.... c5
11. Nb5 O-O { Catastrophic - any reasonable club player can
see this move is a disaster, but Genius2 has no
dynamic knowledge, there is no immediate mate so Genius2
thinks all is ok ! }
12. Rxh6 { CCS2 needs only a few seconds thought to find this move }

bR bN bB bQ -- bR bK --
bP bP -- bN -- bP bP --
-- -- -- -- bP -- -- wR
-- wN bP bP wP -- -- --
-- -- -- wP -- -- -- --
-- -- -- wB -- wN -- --
wP wP wP wQ -- wP wP --
wR -- -- -- wK -- -- --

.... a6 { Incredibly, Genius2 thinks the position is even ! }
13. Bh7+ Kh8
14. Rh5 axb5 { Genius2 still thinks this game is drawn ! }
15. Ke2 { CCS2 finds the killer move .... }
.... Nf6 { Genius2 begins to see the trouble now ... }
16. exf6 Qxf6
17. Rah1 g6
18. Bxg6+ Kg8
19. Rh8+ Qxh8
20. Rxh8+ Kg7
21. Rh7+ Kxg6
22. Qh6+ Kf5 { and mate in 2 more moves. Genius2, the classical
program, soundly defeated by dynamic knowledge.
CCS2 didn't know its attack would win material or
deliver mate, it just knew, dynamically, the the
attack was strong and worth the sacrifice of material. }

This game clearly shows the development and strength of the 'looking-glass'
paradigm. Genius2, a classical program, seemed to have no idea of what
was going on. CCS2 had dynamic knowledge of the strength of its attack from
move 12 on, CCS2 knew from its evaluation function; Genius2 only began to
see the trouble on move 15, seven half-moves later, Genius2's knowledge
was combinational, only 'known' when the search found it.

Who will be the developer ?
===========================
To answer our third question - 'who will be the developer ?', it is necessary to look at the personality of the classical programmers and their hangers-on. These programmers are characterised by a failure to show their emotions (do they ever smile), fear (just watch them operating at tournaments), refusal to discuss how their programs work (just try talking to them) , aversion to taking risks. It has always surprised me that the 'top' programmers are not good chess players. The hangers-on only make a little money, they jealously support their chosen proteges, and viciously attack their opponents. The hangers-on know little, pretend to know much and are governed by fear and greed.
Overall the impression is of a static, non-risk taking, hostile, World War I environment. The new paradigm will come from an unexpected quarter. From a developer with extrovert personality, accustomed to taking risks, a developer with chess knowledge, probably someone unpopular with the classical paradigm supporters, certainly unpopular
with the hangers-on and computer chess entourage. This developer will have been and certainly will be furiously attacked by the classicists.


Search - the lazy programmer's way to avoid evaluating a position.
==================================================================
The new paradigm differs from the classical by one simple conceptual switch.
The classical paradigm makes fast and simple evaluation at each node and generates intelligence from the search tree. The classical programmer looks for ways to make his search more efficient and his evaluation function simpler and faster. The 'looking-glass' paradigm makes slow and complex evaluations at each node and prefers to prune the search tree by use of this evaluation function. In this model search is to be avoided
unless absolutely necessary. Thus the search tree is not central to the new paradigm, rather the search tree is used to find details overlooked, or mistakes made, by the evaluation function. The 'looking-glass' paradigm has the components of human thought - detailed, intuitive evaluation, with search carried out to ensure that the program is not
falling into any traps. I estimate that the difference in nodes per second between and extreme classical program and a 'looking-glass' program will be of the order of 20-30 times, sufficient to give the classical program an extra two plies of search (albeit with reduced knowledge at the nodes). Thus the increased knowledge of the 'looking-glass' program has to compensate for this apparently reduced search depth. The looking-glass strategy necessitates much programming effort, and requires the programmer to have an exceptionally good knowledge of chess strategy and tactics. When such a program is first
being developed it will constantly be outplayed by classical programs, for classical programs see everything within their horizon and the newly developing 'looking-glass' program cannot yet hope to know sufficient tactical and positional themes to compete, but our experience shows that once breakthrough (a knowledge o f sufficient chess themes to compensate for reduced search depth) occurs the looking-glass program begins to
consistently outplay the classical programs. Further advantages emerge from the high level of chess knowledge in the evaluation function - better move selection and move sorting, resulting in more efficient search - more possibilities of accurate forward pruning, resulting in smaller search trees. With increases in tree size (from faster hardware), these advantages are geometric.

B-Search or A-B-Search? - NO! Evaluation based or search based!
===============================================================
The classicists maintain the computer chess dichotomy of B-search (which I understand means pruning occurs at all levels of the tree) or A-B Search (which apparently means that part of the search is full width).
The looking-glass programmer condemns this dichotomy as meaningless.
The new paradigm makes the issue clear: chess programs either have simple evaluation and generate intelligence through search, or have complex evaluations and use limited search as a backup to cover oversights and mistakes. All chess programs prune in one way or another, but looking-glass programs, with complex evaluation, are able to prune more.

Of course, the issue is not so black and white. There is a grey scale between the extreme looking-glass (human play style) and extreme classical style. At the classical end of the scale the B or A-B dichotomy tries to position the program on the scale, but basically classicists believe in search. At the looking-glass end of the scale the issue is how much does the evaluation function allow us to prune or extend - how many risks can we take based on our evaluation function ? Basically looking-glass programmers believe in evaluation.
Von Manstein
============
If, as is said, chess is war, then there must be lessons to be learnt from military history. I have already alluded to the static, boring First World War style of the classical programs (and their programmers !). The opposite style can be found in several histories, Rommel in North Africa, Alexander the Great against Darius, Von Manstein in Russia. Alexander, despite being outnumbered many times, concentrated the powerful mobile part of his army, attacked the stronger Persians, cut through and went straight for Darius himself. The bulk of Darius's army was not engaged, but the battle was decisively won - a classic king attack. Von Manstein (and Rommel) both understood that the power of the outnumbered German army lay in superior staff work, concentration of forces, striking blows to knock the enemy off balance. The looking-glass chess program must contain knowledge of these dynamic elements; and it is only the looking-glass program that has the knowledge and evaluation time available to calculate such ephemerals.

Tal function
============
To find a chess player who understood the king attack, the concentration of forces, the striking of blows to unbalance the opponent, one need look no further than Michael Tal, Russian grandmaster, and player of such romantic and swashbuckling style that his games continue to thrill all lovers of chess. For the developers of the Complete Chess System 2 it was an emotional, and unexpected, experience to find their program playing, sacrificing, in the style of Tal. Opposing programs, well respected, began to fall like dominoes, they appeared to have absolutely no understanding of CCS2's style. We
were almost able to guarantee exciting games against all our opponents.

We believe that the progress we have made with our program, the looking-glass algorithm which we have developed gives us the justification to call our program the Complete Chess System 2 - TAL.

At the time this was all written , we used 486 at 33 MHz !
You have to remember this because today the hardware situation is completely different,

Code: Select all

Dynamic knowledge v. Combinational knowledge
============================================

Oxford Softworks CCS2-v9.0
White: CCS2 486/33
Black: Genius2 486/33
Venue: 1 minute per move
Comment: 1-0

1.  e4  e6
2.  d4  d5 1
3.  Nc3   Nf6 3
4.  Bg5   Be7 5
5.  e5  Nfd7 8
6.  h4  Bxg5
7.  hxg5  { CCS2's opening book ends }
    ....  Qxg5
8.  Nf3   Qd8  { Genius2's opening book ends }
9.  Bd3   h6
10. Qd2   { CCS2's dynamic knowledge - preventing O-O because
    of the threat of Rxh6 }
    ....  c5
11. Nb5   O-O  { Catastrophic - any reasonable club player can
see this move is a disaster, but Genius2 has no
dynamic knowledge, there is no immediate mate so Genius2
thinks all is ok ! }
12. Rxh6  { CCS2 needs only a few seconds thought to find this move }

  bR  bN  bB  bQ  --  bR  bK  --
  bP  bP  --  bN  --  bP  bP  --
  --  --  --  --  bP  --  --  wR
  --  wN  bP  bP  wP  --  --  --
  --  --  --  wP  --  --  --  --
  --  --  --  wB  --  wN  --  --
  wP  wP  wP  wQ  --  wP  wP  --
  wR  --  --  --  wK  --  --  --

    ....  a6   { Incredibly, Genius2 thinks the position is even ! }
13. Bh7+  Kh8
14. Rh5   axb5 { Genius2 still thinks this game is drawn ! }
15. Ke2   { CCS2 finds the killer move .... }
    ....  Nf6  { Genius2 begins to see the trouble now ... }
16. exf6  Qxf6
17. Rah1  g6
18. Bxg6+ Kg8
19. Rh8+  Qxh8
20. Rxh8+ Kg7
21. Rh7+  Kxg6
22. Qh6+  Kf5  { and mate in 2 more moves. Genius2, the classical
program, soundly defeated by dynamic knowledge.
CCS2 didn't know its attack would win material or
deliver mate, it just knew, dynamically, the the
attack was strong and worth the sacrifice of material. }

This game clearly shows the development and strength of the 'looking-glass'
paradigm. Genius2, a classical program, seemed to have no idea of what
was going on. CCS2 had dynamic knowledge of the strength of its attack from
move 12 on, CCS2 knew from its evaluation function; Genius2 only began to
see the trouble on move 15, seven half-moves later, Genius2's knowledge
was combinational, only 'known' when the search found it.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

FWCC
Posts: 117
Joined: Wed Aug 22, 2007 2:39 pm

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by FWCC » Sun Dec 24, 2017 1:18 pm

Thank you Carl for reposting these with the pisitions,I did something wrong positions did not post.

User avatar
Eelco de Groot
Posts: 4172
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: AlphaZero SF game 10 Does Your Engine Find 29Qh3?

Post by Eelco de Groot » Sun Dec 24, 2017 1:36 pm

carldaman wrote:
FWCC wrote:rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29
In this position will your engine find Qh3 which engines say is STILL EQUAL?
I used latest SF and the move was not a choice.Engines like 29Bd4+

This game is amazing to me due to this position:
rn3r2/p3b2p/2p3k1/1p3bp1/8/q1P1B1P1/P4PB1/3RR1KQ b - - 0 26
The Queen is positioned on H1 out of play.This looks like something from science fiction,an Alien game!After going over this game,the whole point was that Black's Queenside was NEVER developed,and here AlphaZero makes the TAL decision to hang the knight for unseen compensation,truly amazing!
rnb2r2/p3bpkp/1ppq3N/6p1/Q7/2P3P1/P4PBP/R1B1R1K1 b - - 0 19
The turning point that I found in the game was 32...Re8 in this position
rn2r1q1/p5k1/2p2bp1/1p4p1/2P5/4B1PQ/5PK1/3R3R w - - 0 33
A truly remarkable game,but does your engine find or like 29.Qh3?
Just checked the first position, I'll look at the others later.

FEN: rn3r2/p3b1k1/2p3p1/1p4p1/8/q1P1B1P1/P4P2/3RR1KQ w - - 0 29

Zero-the-Hero:
1/1 00:00 158 158k +6.17 Qh1-e4 Qa3xa2
2/2 00:00 334 334k +14.32 Qh1-e4 Qa3xc3 Qe4xe7+ Kg7-g8 Be3xg5
3/4 00:00 527 527k +11.36 f2-f4 Qa3xc3 f4xg5
4/6 00:00 1k 1,302k +9.69 Be3-d4+ Be7-f6 Bd4xf6+ Kg7xf6 Qh1-e4 Kf6-g7
6/8 00:00 3k 1,473k +10.60 Be3-d4+ Rf8-f6 Re1-e6 Nb8-d7 Rd1-e1 Qa3xa2 Re6xe7+ Qa2-f7
7/10 00:00 8k 1,944k +8.35 f2-f4 Qa3xc3 f4xg5 Be7xg5 Be3-d4+ Qc3xd4+ Rd1xd4 c6-c5
8/12 00:00 20k 2,171k +6.27 Be3-d4+ Be7-f6 Re1-e3 g5-g4 Qh1-h4 Bf6xd4 c3xd4 Rf8-h8 Re3-e7+ Qa3xe7
20/37 00:00 2,474k 3,475k +0.85 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4 Rd2-e2 Qc5
-f5 Re6-e7+ Rf8-f7 Re7xf7+ Qf5xf7 Qh1-h4 Nb8-d7 Re2-e7 Nd7-e5 Kg1-g2 Kg7-g8 Re7xf7 Kg8xf7 Qh4-h7+ Kf7-e6

21/37 00:00 3,142k 3,415k +1.93 Qh1-h3 Be7-f6 Kg1-g2 Rf8-h8 Qh3-g4 Rh8-h5 Qg4-c8 Qa3xa2 Rd1-d6 Rh5-h8 Qc8-b7+ Qa2
-f7 Qb7xa8 Bf6-e5 Qa8xa7 Qf7xa7 Be3xa7 Be5xd6 Ba7-d4+ Kg7-g8 Re1-e8+ Bd6-f8 Bd4xh8 Nb8-d7 Re8-d8 Kg8xh8 Rd8xd7 Bf8-g7 Rd7
-c7 Bg7xc3 Rc7xc6 b5-b4 Rc6xg6

27/59+ 00:22 65,423k 2,967k +1.85 Qh1-h3
27/60 00:24 74,463k 2,984k +2.78 Be3-d4+ Be7-f6 Bd4xf6+ Rf8xf6 Re1-e8 Rf6-f8 Re8-e6 Qa3-c5 Rd1-d2 g5-g4
Qh1-e4 Qc5-f5 Qe4-d4+ Kg7-g8 Re6-e7 Qf5-f6 Qd4-e3 Nb8-a6 Re7-e6 Qf6-f5 Rd2-d6 Kg8-g7 Qe3-d4+ Kg7-h6 Re6xg6+ Qf5xg6 Rd6xg6+
Kh6xg6 Qd4xg4+ Kg6-f6 Qg4-f3+ Kf6-g7 Qf3xc6 Ra8-d8 Qc6xa6 Rd8-d1+ Kg1-g2 Rd1-d2 Qa6xa7+ Rf8-f7 Qa7-c5 Rd2xf2+ Qc5xf2 Rf7xf2+ Kg2xf2 Kg7-f7 Kf2-f3 Kf7-f6 c3-c4 b5-b4 c4-c5 Kf6-e6 Kf3-e4 Ke6-d7

29/48- 00:38 117,287k 3,062k +1.72 Be3-d4+ Be7-f6
29/49+ 00:39 120,504k 3,074k +1.81 Qh1-h3
33/53- 01:35 322,085k 3,381k +4.23 Qh1-h3 Be7-f6

Cheers,
CL 8-)[/b]
Carl, what on earth is that for engine :o Is it a new program from Ed? Or the new Houdini perhaps?

Image
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

Post Reply