Dylan Sharp Vs. Harvey Williamson (G4)

[Ovyron]

Needless to say, two more Stockfish moves ...

Needless to say, I figured out what Stockfish Depth 70 would play up to 14. Rxc5, in 10 minutes, from an empty tree

You know what's funny? That something happened when I found out we could transpose back to analysis after 11...Bd6 12. Rc1 O-O 13. Nxc5 Bxc5 14. Rxc5 Qd7. I was welcomed with a surprise...

Remember this post of mine from the original thread?

I'm willing to defend 1.g4 against anybody that seriously think it is losing, and I guarantee that I can draw them. The main reason it's ? is because you give any chances of winning whatsoever, so it's only black that can win if you're not careful.

And I'm not saying something like "1.g4 is a draw with perfect play", even at some imperfect playing level it's a draw.

I propose the following experiment:

Use Stockfish at Depth 24 at MultiPV=4 and try to beat it. Write down all its scores and moves for every position it plays. Once you get it to show a -2.00 score, you assume you beat it, then you jump back to the best scored position that it didn't play. You call this a "takeback". The question is: can you beat Stockfish after 24 takebacks?

I estimate that somewhere around takeback 12-16 it'll have found the mainline of 1.g4 (for your attacking line) and no matter what you do, you'll not be able to beat it from there. So you'll have to change your attacking line. My condition is that if you do this, you reset the takebacks back to 24, and try again.

You wouldn't be able to beat Stockfish Depth 24 MultiPV=4 like this (once you make it reach -2.00 and it has ran out of takebacks), so this exercise will help you find and see the drawing moves against the attacking lines you throw at it. This can only happen if 1.g4 is drawn (unfortunately, you can never run out of attacking lines, so you can only prove 1.g4 isn't losing against any line, but never prove it isn't losing against every line.)

Guess what? The moves 11...Bd6 12. Rc1 O-O 13. Nxc5 Bxc5 14. Rxc5 Qd7 transpose back into Stockfish Depth 24 MutliPV=4's line! It was reached via:

1.g4 d5 2.g5 e5 3.d4 exd4 4.Nf3 c5 5.Bg2 Nc6 6.c3 dxc3 7.Nxc3 Nge7 8.O-O d4 9.Ne4 Ng6 10.a3 Bg4 11.Bd2 Qd7 12.Rc1 Be7 13.Nxc5 Bxc5 14.Rxc5 O-O

Does it look familiar?

[d]r4rk1/pp1q1ppp/2n3n1/2R3P1/3p2b1/P4N2/1P1BPPBP/3Q1RK1 w - -

So before the game started I tried (and failed) to beat Stockfish's Depth 24 MultiPV=4 with, what is effectively, Harvey's line!

What are the chances?? I think this should be proof that either Harvey or me or both have been playing the best possible line or one that transposes to it by force. It could also mean that Stockfish Depth 60 isn't any better than Stockfish Depth 24 MultiPV=4 with takebacks (less than 24), and that one gives me a main move and eval after less than a minute!

[zullil]

Needless to say, two more Stockfish moves ...

Needless to say, I figured out what Stockfish Depth 70 would play up to 14. Rxc5, in 10 minutes, from an empty tree

You know what's funny? That something happened when I found out we could transpose back to analysis after 11...Bd6 12. Rc1 O-O 13. Nxc5 Bxc5 14. Rxc5 Qd7. I was welcomed with a surprise...

Remember this post of mine from the original thread?

I'm willing to defend 1.g4 against anybody that seriously think it is losing, and I guarantee that I can draw them. The main reason it's ? is because you give any chances of winning whatsoever, so it's only black that can win if you're not careful.

And I'm not saying something like "1.g4 is a draw with perfect play", even at some imperfect playing level it's a draw.

I propose the following experiment:

Use Stockfish at Depth 24 at MultiPV=4 and try to beat it. Write down all its scores and moves for every position it plays. Once you get it to show a -2.00 score, you assume you beat it, then you jump back to the best scored position that it didn't play. You call this a "takeback". The question is: can you beat Stockfish after 24 takebacks?

I estimate that somewhere around takeback 12-16 it'll have found the mainline of 1.g4 (for your attacking line) and no matter what you do, you'll not be able to beat it from there. So you'll have to change your attacking line. My condition is that if you do this, you reset the takebacks back to 24, and try again.

You wouldn't be able to beat Stockfish Depth 24 MultiPV=4 like this (once you make it reach -2.00 and it has ran out of takebacks), so this exercise will help you find and see the drawing moves against the attacking lines you throw at it. This can only happen if 1.g4 is drawn (unfortunately, you can never run out of attacking lines, so you can only prove 1.g4 isn't losing against any line, but never prove it isn't losing against every line.)

Guess what? The moves 11...Bd6 12. Rc1 O-O 13. Nxc5 Bxc5 14. Rxc5 Qd7 transpose back into Stockfish Depth 24 MutliPV=4's line! It was reached via:

1.g4 d5 2.g5 e5 3.d4 exd4 4.Nf3 c5 5.Bg2 Nc6 6.c3 dxc3 7.Nxc3 Nge7 8.O-O d4 9.Ne4 Ng6 10.a3 Bg4 11.Bd2 Qd7 12.Rc1 Be7 13.Nxc5 Bxc5 14.Rxc5 O-O

Does it look familiar?

[d]r4rk1/pp1q1ppp/2n3n1/2R3P1/3p2b1/P4N2/1P1BPPBP/3Q1RK1 w - -

So before the game started I tried (and failed) to beat Stockfish's Depth 24 MultiPV=4 with, what is effectively, Harvey's line!

What are the chances?? I think this should be proof that either Harvey or me or both have been playing the best possible line or one that transposes to it by force. It could also mean that Stockfish Depth 60 isn't any better than Stockfish Depth 24 MultiPV=4 with takebacks (less than 24), and that one gives me a main move and eval after less than a minute!

Do you understand that my Stockfish is reaching high depth only because you are playing at correspondence time control?

In any case, so far you are perfect, according to Stockfish. But then, that's not surprising, since you're getting your moves from Stockfish.

[Ovyron]

Do you understand that my Stockfish is reaching high depth only because you are playing at correspondence time control?

Do you understand your high depth Stockfish isn't producing anything better than what Stockfish Depth 24 at MultiPV=4 was showing 4 days before the game started?

So what exactly does one get by getting a 20 core machine like yours? What does one get by letting it reach Depth 60? What are you demonstrating by telling us its eval at high depth?

I guess we'll have to see what its 11th move choice was if it wasn't 11...Qd7, if its -1.64 eval was accurate and black isn't better now it had in its PV a better move than what Harvey or me could see, and that'd be something. If not then the eval it's showing isn't worth all the resources you're throwing at it.

(for the record, after I accept Harvey's conditionals we'll be back to the lines where I can guarantee a draw, though I guess that doesn't matter now)

[zullil]

Do you understand that my Stockfish is reaching high depth only because you are playing at correspondence time control?

Do you understand your high depth Stockfish isn't producing anything better than what Stockfish Depth 24 at MultiPV=4 was showing 4 days before the game started?

So what exactly does one get by getting a 20 core machine like yours? What does one get by letting it reach Depth 60? What are you demonstrating by telling us its eval at high depth?

I guess we'll have to see what its 11th move choice was if it wasn't 11...Qd7, if its -1.64 eval was accurate and black isn't better now it had in its PV a better move than what Harvey or me could see, and that'd be something. If not then the eval it's showing isn't worth all the resources you're throwing at it.

(for the record, after I accept Harvey's conditionals we'll be back to the lines where I can guarantee a draw, though I guess that doesn't matter now)

Discussion between us seems pointless.

What is interesting is that this experiment further convinces me that Stockfish's most significant weakness is in its static evaluation of near tablebase positions. This is increasingly important, because on decent hardware Stockfish will be reaching such positions from very early in the game, and will make choices that steer the game into (or away from) certain such positions. When a drawn 8-man position is given a static evaluation of 1.5 (for example), Stockfish might well head there, instead of heading for a winning endgame with a static eval of, say, 1.3.

Certainly expert human guidance can help here, and I suspect that many of the (apparently, rare) decisive games in high-level correspondence play result from such human expertise.

In any case, I've been away from my computer for a while, so I'm just now about to enter Rc1 and O-O. The eval before I enter them is currently -1.54 at depth 73.

[EDIT] Just entered both moves. After 1 minute, Stockfish is at depth 47.

[Ovyron]

Stockfish might well head there, instead of heading for a winning endgame with a static eval of, say, 1.3.

For that you need to prove that such a position exists. Can you show an endgame position where Stockfish shows an 1.3 and it's winning for white? Because maybe all can be defended, so with the so called "best play", an endgame position might be 0.00 or +2.00 (winning.)

[Raphexon]

I'm doubtful SF often changes its mind beyond depth 45-50 or so.

[Zenmastur]

Discussion between us seems pointless.

What is interesting is that this experiment further convinces me that Stockfish's most significant weakness is in its static evaluation of near tablebase positions. This is increasingly important, because on decent hardware Stockfish will be reaching such positions from very early in the game, and will make choices that steer the game into (or away from) certain such positions. When a drawn 8-man position is given a static evaluation of 1.5 (for example), Stockfish might well head there, instead of heading for a winning endgame with a static eval of, say, 1.3.

Certainly expert human guidance can help here, and I suspect that many of the (apparently, rare) decisive games in high-level correspondence play result from such human expertise.

So you are learning something. Imagine that!

You are very much right about the endgames influence on early positions in CC. You're also right about SF's eval being an impediment to selecting the right course of action. When it comes to endgame evaluations SF can be a complete nit wit. You will spend half your time trying to convince SF of something YOU already know! But if you want it's help you have to spend the time to get it to realize the “true” value of those positions otherwise it's analysis will be flawed by it's evaluation function.

You are, however, wrong to conclude that wins come from human endgame expertise. It turn's out that many times the players are blinded by the engines single number evaluations. It turns out that this is a less than ideal way to evaluate a chess position.

Regards,

Zenmastur

[zullil]

Stockfish might well head there, instead of heading for a winning endgame with a static eval of, say, 1.3.

For that you need to prove that such a position exists. Can you show an endgame position where Stockfish shows an 1.3 and it's winning for white? Because maybe all can be defended, so with the so called "best play", an endgame position might be 0.00 or +2.00 (winning.)

Must be lots and lots of such examples. Here's one where White wins but Stockfish's static eval favors Black. Assuming I interpreting the eval properly.

Code: Select all

Stockfish 191019 64 BMI2 by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
position fen 5r2/8/8/8/3K4/PPP4P/8/7k w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   | r |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   | K |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 | P | P | P |   |   |   |   | P |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | k |
 +---+---+---+---+---+---+---+---+

Fen: 5r2/8/8/8/3K4/PPP4P/8/7k w - - 0 1
Key: E57C7BFFF6692166
Checkers: 
eval
     Term    |    White    |    Black    |    Total   
             |   MG    EG  |   MG    EG  |   MG    EG 
 ------------+-------------+-------------+------------
    Material |  ----  ---- |  ----  ---- | -3.54 -1.81
   Imbalance |  ----  ---- |  ----  ---- |  0.19  0.19
       Pawns |  0.25 -0.20 |  0.00  0.00 |  0.25 -0.20
     Knights |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
     Bishops |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Rooks |  0.00  0.00 |  0.22  0.12 | -0.22 -0.12
      Queens |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
    Mobility |  0.00  0.00 |  0.27  0.80 | -0.27 -0.80
 King safety | -0.62 -0.14 | -0.93  0.02 |  0.32 -0.16
     Threats |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
      Passed |  0.16  0.51 |  0.00  0.00 |  0.16  0.51
       Space |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
  Initiative |  ----  ---- |  ----  ---- |  0.00 -0.02
 ------------+-------------+-------------+------------
       Total |  ----  ---- |  ----  ---- | -3.10 -2.41

Total evaluation: -1.34 (white side)

[Guenther]

I'm doubtful SF often changes its mind beyond depth 45-50 or so.

And what? We have >65 depths here from Louis since the beginning.

[Ovyron]

Oops, I missed the word "static", I was talking about the eval that you'd get after searching