AlphaZero - Stockfish game 3

[Ovyron]

Again the pointless exercise of analyzing with the losing Engine, Stockfish.

Well, if you go to the core of it, analyzing chess positions with whatever you have available is a pointless excercise, because any state of the art analysis you make will be obsolete down the road.

This didn't stop people from analyzing their games 9 years ago with Rybka 3, back then it was the strongest thing available, and it was crushing the second strongest thing 3 times harder than A0 defeated Stockfish.

Nowadays you can take a look at Rybka 3's analysis and say it's garbage compared to Stockfish. However, if you had access to A0's analysis it wouldn't change a thing, unless the position was easy, you'd be just as far away from perfection as you're now.

A0 is just a new paradigm, it's like the Rybka 1.0 Beta of our time, that back then kicked something up that everyone did, and skyrocketted chess engine strength by 600 ELO. Can it happen again?

Can you imagine a chess entity that can defeat Stockfish by 600 ELO instead of 100? We could have it in 10 years, and then any analysis you could have made with A0 would look like garbage, making analyzing with it a pointless exercise.

Stockfish drew 73 of its games on the match against it, after all, so in those games 1 minute pointless moves sufficed to stop the beast.

[TommyTC]

Again the pointless exercise of analyzing with the losing Engine, Stockfish.
What makes you think that AlphaZero would be so obliging as going along with Stockfish's "Analysis" ?

I am not using the "losing" engine, and I'm not using it to "play" against another computer at one minute per move. I am using the latest SF with full 6 piece EGTB and I'm analyzing positions and variations resulting from those positions. The positions I'm analyzing are late middle game and endgame positions, and for this SF is not worse than AlphaZero.

The art of analyzing a chess position to arrive at an objective conclusion of a position has nothing to do with the opponent being better (or worse), so it does not matter what AlphaZero would "play" from any of the positions. With proper analysis, using any of the current top engines, it is possible to arrive at conclusions about many of the positions arising after the move 49. Rf6. Proper analysis of a position is not just setting up 1 position and letting an engine run for hours and then looking at it's evaluation and PV.

To be clear, earlier someone posted that 49... Kf8 would hold the position for Stockfish (Black). I found an alternative move (54. Qc3) in a variation which casts doubt on it being so clearly drawn.

Oh, and by the way, what was the evaluation of AlphaZero after 49. Rf6? And what was its principal variation?

[TommyTC]

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

45/67 80:19 +0.67 49...Kf8 50.Qf4 Qg8 51.Qc7 c4
52.Qxc4 Rd8 53.Qd4 Ke8 54.g4 Qf8
55.g5 Re6 56.Rf3 a5 57.Bd5 Qb4
58.Qg7 Qf8 59.Qxh7 Ke7 60.Kh2 Rb8
61.Bc4 Rc8 62.Bxe6 dxe6 (11.005.092.745) 2283

I'm wondering if it is an improvement in this last PV to have the white queen on c3 instead of d4:

49...Kf8 50. Qf4 Qg8 51. Qc7 c4 52. Qxc4 Rd8 53. g4 Ke8 54. Qc3 Qf8 55. g5 Re6 56. Rf3

I've had SF analyze to depth 70 and evaluate this as +1.75. There are a lot of tricky R+P endgames to evaluate. I'm not convinced either way if it's winning or drawing.

Hi Thomas, this is what I get with Kaissa after 54. Qc3

3rk1q1/p2prp1p/5RpP/8/6P1/1BQ5/P4PK1/8 b - -

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



52/77 1004:40+0.62 54...Qf8 55.g5 Re6 56.Rf3 Qe7
57.Qh8+ Qf8 58.Qxh7 Ke7 59.Bxe6 dxe6
60.Rc3 Rd5 61.Rc7+ Kd6 62.Qxf7 Rxg5+
63.Kf1 Qxf7 64.Rxf7 Rh5 65.h7 a5
66.Ra7 e5 67.Ra6+ Kc5
{[D]8/7P/R5p1/p1k1p2r/8/8/P4P2/5K2 w - -} (163.599.434.339) 2713



Leaf:
[pgn]
[Event "?"]
[Site "?"]
[Date "2017.12.17"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "8/7P/R5p1/p1k1p2r/8/8/P4P2/5K2 w - -"]

1. Rxa5+ Kb6 2. Rxe5 Rxh7 *
[/pgn]

We are a long way from the starting position of this discussion.

I have the full 6 piece EGTB on SSD. From the diagram position after 68. Rxa5 Kb6 69. Rxe5 Rxh7 70. Re6 Kc5 71. Rxg6 it is tablebase draw. Moves other than 71. Rxg6 result in evaluation around +0.66 to depth 81. I think it is safe to conclude this leaf position is drawn.

Walking back this variation I've found the evaluation does not change from move 60. Rc3 to the end of the variation. I've analyzed beyond depth 50 and the eval stays below +1.00, so from this move forward I believe the position is drawn.

I think 60. Ra3, 60. Rb3, and 60. Rc3 are all drawn.
But even at move 60 we are already 20 ply from the position after 49...Kf8. I don't know if there is an improvement earlier in these variations.

[IQ]

But even at move 60 we are already 20 ply from the position after 49...Kf8. I don't know if there is an improvement earlier in these variations.

People! What are you doing here? Alphazero's games should have taught you one lesson and one lesson only: Do not trust SF and think for yourself! SF analysis in this position is completely misguided.

Ask yourself what is important in this position? The c5 and/or d7 pawns? No! It's the a7 pawn, if white wins that pawn his own a-pawn becomes unstoppable if supported by a bishop on d5.

Your lines of analysis after Kf8 Qf4 and especially the later g4 g5 remind be of a poor mans alpha parody which seems to scream "Hey see I can suffocate my opponents, too" except NO you can't. Not only is the c5 pawn not that important but g4 and g5 weaken whites kingside too much and allow black counterplay in those rook endings.

I propose the following human play after 49. ... Ke8. Instead of 50. Qf4 you should look at:
50. Bd5! Qg8
51. Qa4! +-
51. ... Rd8 (if 51. ... d6 to protect a7 then all the Re6 defense ideas fail for black and white wins easily and if one sneakily tries to distract whites queen from a7 with 51. ... Qh8 then after Qf4! Qg8 Qc7 Qh8 Qd6! - watch the pathetic black Queen - black also wins easily as he won 2 important tempi. SF is utterly confused here probably due to null move pruning)
52. Qxa7 Qh8 (what else as Ke8 looses another pawn)
53. Qa5 Ke8
54. Qc3 Qf8
55. a4 +- and the position plays itself

[chessmobile]

But even at move 60 we are already 20 ply from the position after 49...Kf8. I don't know if there is an improvement earlier in these variations.

People! What are you doing here? Alphazero's games should have taught you one lesson and one lesson only: Do not trust SF and think for yourself! SF analysis in this position is completely misguided.

Ask yourself what is important in this position? The c5 and/or d7 pawns? No! It's the a7 pawn, if white wins that pawn his own a-pawn becomes unstoppable if supported by a bishop on d5.

Your lines of analysis after Kf8 Qf4 and especially the later g4 g5 remind be of a poor mans alpha parody which seems to scream "Hey see I can suffocate my opponents, too" except NO you can't. Not only is the c5 pawn not that important but g4 and g5 weaken whites kingside too much and allow black counterplay in those rook endings.

I propose the following human play after 49. ... Ke8. Instead of 50. Qf4 you should look at:
50. Bd5! Qg8
51. Qa4! +-
51. ... Rd8 (if 51. ... d6 to protect a7 then all the Re6 defense ideas fail for black and white wins easily and if one sneakily tries to distract whites queen from a7 with 51. ... Qh8 then after Qf4! Qg8 Qc7 Qh8 Qd6! - watch the pathetic black Queen - black also wins easily as he won 2 important tempi. SF is utterly confused here probably due to null move pruning)
52. Qxa7 Qh8 (what else as Ke8 looses another pawn)
53. Qa5 Ke8
54. Qc3 Qf8
55. a4 +- and the position plays itself

What happens after 55..Qxh6. Have you tried to win this position with the help of any software you like against Stockfish set to a fair time control. I would love to see A0 resume this game at a future date after SF is given the chance to play 49..Kf8.

[shrapnel]

People! What are you doing here? Alphazero's games should have taught you one lesson and one lesson only: Do not trust SF and think for yourself! SF analysis in this position is completely misguided.

You're wasting your Breath trying to convince these people who have their Minds set in stone.
You might as well wring blood from a stone than try convince these Golden Oldies to accept any new ideas.
They run on a railway track called Alpha-Beta with no possible deviations.

[IQ]

What happens after 55..Qxh6. Have you tried to win this position with the help of any software you like against Stockfish set to a fair time control. I would love to see A0 resume this game at a future date after SF is given the chance to play 49..Kf8.

Just follow my plan.

55. ... Qxh6
56. Rf3 f5 (probably only move)
57. a5 +/-

please just think for yourself, or alternatively just play through the moves and do not just sit there and let SF run. At the end white is a piece up most of the times and retains a white pawn, which is enough to win. If you have a miracle defense, please just post it here.

[Eelco de Groot]

People! What are you doing here? Alphazero's games should have taught you one lesson and one lesson only: Do not trust SF and think for yourself! SF analysis in this position is completely misguided.

You're wasting your Breath trying to convince these people who have their Minds set in stone.
You might as well wring blood from a stone than try convince these Golden Oldies to accept any new ideas.
They run on a railway track called Alpha-Beta with no possible deviations.

Fanboys are of all ages

[Eelco de Groot]

What happens after 55..Qxh6. Have you tried to win this position with the help of any software you like against Stockfish set to a fair time control. I would love to see A0 resume this game at a future date after SF is given the chance to play 49..Kf8.

Just follow my plan.

55. ... Qxh6
56. Rf3 f5 (probably only move)
57. a5 +/-

please just think for yourself, or alternatively just play through the moves and do not just sit there and let SF run. At the end white is a piece up most of the times and retains a white pawn, which is enough to win. If you have a miracle defense, please just post it here.

It was a nice find Roberto. Kaissa thinks it might just be enough too. Full Bishop against two pawns. But a piece is a piece. I think I saw 50...Qg8 is not totally forced at the beginning (not Sf's first choice) of your variation so you'd have to investigate the alternatives there but that might not make a difference.

This is the output from Kaissa after 55...Qxh6, two best:

[D]3rk3/3prp1p/5Rpq/2pB4/P7/2Q3P1/5PK1/8 w - -

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

49 1004:19+1.17 56.Rf3 f5 57.a5 d6 58.Bc6+ Kf7 59.a6 Qg7
60.Qa5 Qf8 61.g4 Re5 62.a7 Kg7
63.Qa1 h6 64.Bd5 Kh7 65.a8Q Rxa8
66.Bxa8 Re7 67.g5 Qh8 68.Qa5 f4
69.Rxf4 {[D]B6q/4r2k/3p2pp/Q1p3P1/5R2/8/5PK1/8 b - -}(144.171.256.159) 2392

49 1004:19+0.08 56.Rb6 Re6 57.Bxe6 dxe6 58.Rb7 Rd7
59.Rb8+ Ke7 60.Qxc5+ Kf6 61.Qc3+ Ke7
62.Qa3+ Kf6 63.Qf3+ Ke7 64.a5 Qd2
65.a6 Qd6 66.Qa8 Qc7 67.Rb3 Qc4
68.Ra3 Rd8 69.Qf3 (144.171.256.159) 2392

Leafposition:

B6q/4r2k/3p2pp/Q1p3P1/5R2/8/5PK1/8 b - -

Engine: Kaissa (64 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott


40/59 10:13 +1.46 1...Qe5 2.Rg4 h5 3.Rg3 Rf7 4.Qa2 Rf5
5.Qa7+ Qg7 6.Qxg7+ Kxg7 7.Be4 Re5
8.Bd3 Rd5 9.Bb5 Rf5 10.Bd7 Rd5 11.f4 Rd2+
12.Kh3 c4 13.Bc6 Rd4 14.Rf3 c3 (1.388.020.364) 2261

[IQ]

It was a nice find Roberto. Kaissa thinks it might just be enough too. Full Bishop against two pawns. But a piece is a piece. I think I saw 50...Qg8 is not totally forced at the beginning (not Sf's first choice) of your variation so you'd have to investigate the alternatives there but that might not make a difference.

Yes, i have my moments. Especially since it has become so rare to find something which SF does not see after while. Kudos to alpha for pressuring SF into such positions in the first place.

A couple of notes on your kaissa lines (kaissa and SF unfortunately fall into a number of traps):
1) Do not play 63. Qa1. This could lead to a number of rook + pawns vs. queen fortresses that SF just does not understand - SF scores them highly but they are draw. Instead just promote the pawn and win the resulting endings (all won).
2) Yes there is one serious alternative to Qg8 at the beginning but it is also won accordingly with similar play.