1.g4 opening is losing?

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

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
jp
Posts: 1442
Joined: Mon Apr 23, 2018 5:54 am

Re: 1.g4 opening is losing?

Post by jp » Mon Sep 28, 2020 4:30 am

Alayan wrote:
Sun Sep 27, 2020 5:43 pm
The only true values of a chess position are draw and mate in X. A score like 150cp is always incorrect from a chess truth perspective, it's only a useful approximation for humans and engines to denote a position that's very difficult to hold, maybe losing by force, maybe not. So of course it can't be stable.
Evals can be stable in the sense that engines stick to a line and as the depth goes up the eval is stable. That gives us confidence. If as the depth changes, or for different engines, the eval is bouncing around between negative and positive values (for example), then that's what I mean by "unstable" and untrustworthy.

Of course, if an engine eval goes from +3 at depth 40 to +5 at depth 60 and +10 at depth 80, maybe I would not call the eval "stable", but the direction of the change is stable.

Alayan
Posts: 529
Joined: Tue Nov 19, 2019 7:48 pm
Full name: Alayan Feh

Re: 1.g4 opening is losing?

Post by Alayan » Mon Sep 28, 2020 3:26 pm

Stability of an engine eval from depth to depth is achieved by requiring a new best move to be searched to the same depth as the previous best move. In ChessDB, you'll get the minimum leaf eval of 22, but otherwise a tree with a handful of nodes can be backed up to the root and through minimaxing take preference over a tree with a million nodes. Automatic exploration will later pick it up and expand the small tree and either refute a key move (and go back to the big tree) or confirm the findings and keep it.

This has some advantages, because the moves may be explored to very unequal depths that don't reflect their quality, and the search tree shape is less regular than in an engine search. If you could just explore the line 100 plies deep and not have the root score change to another line until it is also explored 100 plies deep, it would be an issue.

But it comes with the drawback that the root score is easier to change with a shallow line.

In practice, black must walk a fine line to keep the (likely) winning advantage, while white has a lot of drawing attempts available. The difficulty to move the score is directly related to the number of lines that must get scored differently, and this number of lines is very roughly (number of enemy moves to refute)^(depth to refute).

If you were to attempt to change CDB's 1. g4 score, you'll manage a much bigger change by showing it a way to draw what it thought was a key attacking line than by showing it a way to bust a drawing attempt. On the other hand, once you've pushed the score down, your drawing line will be the only one with such a low score to refute through a new attacking move to get back to the previous evaluation.

So the challenge for someone that wants to prove a draw is to push the score down and keep it down.

I'm waiting on mmt's big hardware analysis of the 2. c4 Bxg4 line. :D

mmt
Posts: 343
Joined: Sun Aug 25, 2019 6:33 am
Full name: .

Re: 1.g4 opening is losing?

Post by mmt » Fri Oct 02, 2020 3:10 am

After 1.g4 d5 2.c4 Bxg4

Stockfish_20091717_x64_avx2, nn-59da7a08a5d7.nnue, 450 GB cache, most useful 7-piece EGTBs, around 14 trillion nodes, SF or Arena stopped updating the count after 7 trillion:
74/94 -1.23
1.Bg2 c6 2.Qb3 e6 3.Qxb7 Nd7 4.Nc3 Ngf6 5.Qxc6 Rc8 6.Qa6 dxc4 7.Nf3 Bb4 8.d4 cxd3 9.Qxd3 Bf5 10.Qb5 Bxc3+ 11.bxc3 0-0 12.0-0 Nd5 13.Ba3 Re8 14.Rfd1 a6 15.Qb7 N7b6 16.Bb4 Bc2 17.Rd2 Qf6 18.Qxa6 Be4 19.Ne1 Ra8 20.Qb5 Na4 21.Rc1 Bxg2 22.Nxg2 Naxc3 23.Bxc3 Nxc3 24.Qb6 Nxa2 25.Re1 Red8 26.Rxd8+ Rxd8 27.Kf1 Nc3 28.Rc1 Nd5 29.Qa5 Re8 30.Qa3 Qh6 31.h4 Qf6 32.e4 Ne7 33.Qc3 e5 34.f4 Ng6 35.f5 Nxh4 36.Qc6 Qxc6 37.Rxc6 Nf3 38.Kf2 Nd4

The best move for black according to Chessdb.cn is 6... Ne7 not 6... Ngf6. SF NNUE analysis after 2.8 trillion nodes gives it:

72/93 -1.32
1.cxd5 exd5 2.d4 Rb8 3.Qa6 Rb6 4.Qd3 Ng6 5.h3 Be6 6.Nf3 Bd6 7.h4 h6 8.Bh3 Bxh3 9.Rxh3 0-0 10.Rh1 Qf6 11.h5 Ne7 12.b3 c5 13.dxc5 Nxc5 14.Qd4 Qf5 15.Kf1 Rc8 16.Bb2 Ne6 17.Qd3 Qxd3 18.exd3 Nf4 19.Rd1 Kh7 20.Ne2 Rc2 21.Rd2 Rxd2 22.Nxd2 Nxd3 23.Bd4 Rb7 24.Rh3 Nb4 25.a4 Nec6 26.Bc3 Bc5 27.Nf3 Nc2 28.Ne5 Nxe5 29.Bxe5 d4 30.Rg3 f6 31.Bf4 Nb4 32.Nc1 Nc6 33.Ke1 Ba3 34.Kd1 a5 35.Bd2 Bxc1 36.Bxc1 Rd7 37.Ba3 Rd5 38.Rh3 Re5 39.Kd2 Rg5 40.Kc2

Continuing from there, Chessdb.cn agrees with this line but has no analysis at move 19 where it counts it as -2.31, so significantly worse than what I get. I'll analyze this position to make sure.


Since 2... e5 that was analyzed previously was -1.50, so far it still seems like a preferred continuation for black over 2... Bxg4 to me but it's far from settled.

mmt
Posts: 343
Joined: Sun Aug 25, 2019 6:33 am
Full name: .

Re: 1.g4 opening is losing?

Post by mmt » Sat Oct 03, 2020 2:52 am

From this latest position updated Stockfish_20092822_x64_avx2 with nn-baeb9ef2d183.nnue goes to -1.99 at depth 73 and stays at -1.99 to at least depth 95 (1.5 trillion nodes) so it looks drawish:

95/133 -1.99 1.dxc5 Nxc5 2.Qd4 Qf5 3.Kf1 Rfb8 4.Be3 Rb4 5.Qd2 Ne4 6.Nxe4 Rxe4 7.Qd3 a5 8.Bd2 Bc7 9.Rc1 Bb6 10.Qa6 Qf6 11.Bxa5 Bxa5 12.Qxa5 Re8 13.Qb5 Nf5 14.Rc2 Nd4 15.Nxd4 Qxd4 16.Rc1 Qf6 17.Qd3 Qb2 18.Re1 Qxa2 19.b4 Rxb4 20.Rg1 Re5 21.Qc3 Qb2 22.Qxb2 Rxb2 23.Rh1 d4 24.Rh4 Rd2 25.Rh3 Rf5 26.Rh2 d3 27.exd3 Rxd3 28.Re8+ Kh7 29.Re4 Rb3 30.Reh4 Kg8 31.Kg2 Rg5+ 32.Kf1 Rb1+ 33.Ke2 Re5+ 34.Kf3 Ra5 35.Kg2 Rg5+ 36.Kf3 Kh7 37.Ke3 Rb3+ 38.Ke2 Rb2+ 39.Kf3 Rbb5 40.R4h3 Rb1 41.Ke3 Rb3+ 42.Ke2 Re5+ 43.Kf1 Rb1+ 44.Kg2 Rb2 45.Rh4 Rf5 46.Kf1 Rc5 47.Kg2 Re5 48.Kg3 Rb3+ 49.Kg2 Rg5+ 50.Kf1 Kg8 51.R4h3 Rb1+ 52.Ke2 Rb2+ 53.Ke3 Rb3+ 54.Ke2 Re5+ 55.Kf1 Rb1+ 56.Kg2 Rg5+ 57.Kf3 Ra1 58.Ke2 Re5+ 59.Kf3 Rf5+ 60.Kg4 Rg5+ 61.Kf3 Ra4 62.Rh4 Rf5+ 63.Kg2 Ra2 64.Kg3 Ra3+ 65.Kg2 Rg5+ 66.Kf1

Seems that Chessdb.cn cannot account for a stable eval like this where there is a seeming advantage but no real progress? If it continues to just rely on engine scores, these types of positions will have wrong evals, making the whole up-root evaluation suspect.

Maybe Ovyron or somebody else can suggest a better continuation for black somewhere that we can analyze. Or we can play out a bunch of computer vs computer variations to find potential winning black moves for deeper analysis.

Zenmastur
Posts: 919
Joined: Sat May 31, 2014 6:28 am

Re: 1.g4 opening is losing?

Post by Zenmastur » Sat Oct 03, 2020 4:34 am

mmt wrote:
Sat Oct 03, 2020 2:52 am
From this latest position updated Stockfish_20092822_x64_avx2 with nn-baeb9ef2d183.nnue goes to -1.99 at depth 73 and stays at -1.99 to at least depth 95 (1.5 trillion nodes) so it looks drawish:

95/133 -1.99 1.dxc5 Nxc5 2.Qd4 Qf5 3.Kf1 Rfb8 4.Be3 Rb4 5.Qd2 Ne4 6.Nxe4 Rxe4 7.Qd3 a5 8.Bd2 Bc7 9.Rc1 Bb6 10.Qa6 Qf6 11.Bxa5 Bxa5 12.Qxa5 Re8 13.Qb5 Nf5 14.Rc2 Nd4 15.Nxd4 Qxd4 16.Rc1 Qf6 17.Qd3 Qb2 18.Re1 Qxa2 19.b4 Rxb4 20.Rg1 Re5 21.Qc3 Qb2 22.Qxb2 Rxb2 23.Rh1 d4 24.Rh4 Rd2 25.Rh3 Rf5 26.Rh2 d3 27.exd3 Rxd3 28.Re8+ Kh7 29.Re4 Rb3 30.Reh4 Kg8 31.Kg2 Rg5+ 32.Kf1 Rb1+ 33.Ke2 Re5+ 34.Kf3 Ra5 35.Kg2 Rg5+ 36.Kf3 Kh7 37.Ke3 Rb3+ 38.Ke2 Rb2+ 39.Kf3 Rbb5 40.R4h3 Rb1 41.Ke3 Rb3+ 42.Ke2 Re5+ 43.Kf1 Rb1+ 44.Kg2 Rb2 45.Rh4 Rf5 46.Kf1 Rc5 47.Kg2 Re5 48.Kg3 Rb3+ 49.Kg2 Rg5+ 50.Kf1 Kg8 51.R4h3 Rb1+ 52.Ke2 Rb2+ 53.Ke3 Rb3+ 54.Ke2 Re5+ 55.Kf1 Rb1+ 56.Kg2 Rg5+ 57.Kf3 Ra1 58.Ke2 Re5+ 59.Kf3 Rf5+ 60.Kg4 Rg5+ 61.Kf3 Ra4 62.Rh4 Rf5+ 63.Kg2 Ra2 64.Kg3 Ra3+ 65.Kg2 Rg5+ 66.Kf1

Seems that Chessdb.cn cannot account for a stable eval like this where there is a seeming advantage but no real progress? If it continues to just rely on engine scores, these types of positions will have wrong evals, making the whole up-root evaluation suspect.

Maybe Ovyron or somebody else can suggest a better continuation for black somewhere that we can analyze. Or we can play out a bunch of computer vs computer variations to find potential winning black moves for deeper analysis.
Not sure from which position you want a move for but in the line given above the move 26. ..d3 is a blunder. 26. ..f6 looks winning with an eval of <-3.5. The most probable outcome of a full reverse analysis of the given line will be that at some point prior to blacks 26th move the entire line will change and so will it's ending evaluation.

The problem with very long searches (1.5 Trillon nodes) is that unless you have a very large TT (i.e. at least equal to the nodes searched in bytes) they become unreliable simply because during a part of the sub-search the TT will be effectively completely over-written. Things found in one iteration may not/ likely won't be found in the TT during the next iteration and can even be missed by the search. In the case of your search you need at least 1.5TB of TT IMHO.

If you don't have that much TT the only reliable way that I know of is to do a full reverse analysis with as much TT as you have. In this case you need at least 128GB of TT to properly do the reverse searches needed to get a "proper" evaluation to the root.
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

mmt
Posts: 343
Joined: Sun Aug 25, 2019 6:33 am
Full name: .

Re: 1.g4 opening is losing?

Post by mmt » Sat Oct 03, 2020 5:19 am

Zenmastur wrote:
Sat Oct 03, 2020 4:34 am
The problem with very long searches (1.5 Trillon nodes) is that unless you have a very large TT (i.e. at least equal to the nodes searched in bytes) they become unreliable simply because during a part of the sub-search the TT will be effectively completely over-written. Things found in one iteration may not/ likely won't be found in the TT during the next iteration and can even be missed by the search. In the case of your search you need at least 1.5TB of TT IMHO.

If you don't have that much TT the only reliable way that I know of is to do a full reverse analysis with as much TT as you have. In this case you need at least 128GB of TT to properly do the reverse searches needed to get a "proper" evaluation to the root.
Hmm I don't really know how SF does this but shouldn't a subtree simply be evaluated again if it's not in the TT unless there is a hash collision? And shouldn't nodes higher up the tree take precedence, leaving just smaller subtrees that are quick to re-evaluate? This search was run with 450 GB of TT. I'll take a look at f6, thanks.

Zenmastur
Posts: 919
Joined: Sat May 31, 2014 6:28 am

Re: 1.g4 opening is losing?

Post by Zenmastur » Mon Oct 05, 2020 5:12 pm

mmt wrote:
Sat Oct 03, 2020 5:19 am
Zenmastur wrote:
Sat Oct 03, 2020 4:34 am
The problem with very long searches (1.5 Trillon nodes) is that unless you have a very large TT (i.e. at least equal to the nodes searched in bytes) they become unreliable simply because during a part of the sub-search the TT will be effectively completely over-written. Things found in one iteration may not/ likely won't be found in the TT during the next iteration and can even be missed by the search. In the case of your search you need at least 1.5TB of TT IMHO.

If you don't have that much TT the only reliable way that I know of is to do a full reverse analysis with as much TT as you have. In this case you need at least 128GB of TT to properly do the reverse searches needed to get a "proper" evaluation to the root.
Hmm I don't really know how SF does this but shouldn't a subtree simply be evaluated again if it's not in the TT unless there is a hash collision? And shouldn't nodes higher up the tree take precedence, leaving just smaller subtrees that are quick to re-evaluate? This search was run with 450 GB of TT. I'll take a look at f6, thanks.
Well... the problem with this theory is that much of what is analyzed depends on the contents in the TT. If the TT contents are completely different from one iteration to the next it's very unlikely that a highly threaded search will search the same lines. The search isn't very deterministic unless only one thread is used. Even minor differences in the TT contents can lead to a subtree search that is completely different.

The other factor that comes into play is that the move chosen has a lot to do with the search bounds. If you use different bounds, as might happen, if the TT has changed or the entry is no longer available you will very likely end up with a different sub-tree searched, a different line of play given, and a different evaluation. At least this is true if the line of play is unforced. Forcing lines tend not to change much regardless of the TT since one/or both sides have a "clearly" best move.

Since you've had a few days to look at the position, I'm wondering what you came up with.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

mmt
Posts: 343
Joined: Sun Aug 25, 2019 6:33 am
Full name: .

Re: 1.g4 opening is losing?

Post by mmt » Tue Oct 06, 2020 1:40 am

Zenmastur wrote:
Mon Oct 05, 2020 5:12 pm
Since you've had a few days to look at the position, I'm wondering what you came up with.
Something unusual: the eval was -1.26 from depth 21 to depth 82 and it took just 34 minutes to get there. But now it's been searching depth 83 for a very long time:



82/101 34:03 59,120,134k 28,928k -1.26 1.Ra1 Re5 2.Re1 Rd5 3.Rh3 Kh7 4.Rh4 Re5 5.Rh1 d3 6.exd3 Rf5 7.Rh2 Rxd3 8.Re4 Rfd5 9.Ke2 Ra3 10.Kf1 Raa5 11.Reh4 Kg8 12.R2h3 Rg5 13.Rh2 Rac5 14.R4h3 Rce5 15.Rh4 Ra5 16.R2h3 Ra2 17.Re3 Ra1+ 18.Ke2 Raa5 19.Reh3 Rgf5 20.Rh2 Rad5 21.Ke3 Rfe5+ 22.Kf3 Rd3+ 23.Kg2 Rg5+ 24.Kf1 Rd1+ 25.Ke2 Rgd5 26.R2h3 R1d2+ 27.Kf1 Rf5 28.Rh2 Rdd5 29.R2h3 Rc5 30.Kg2 Rc1 31.Rh2 Rc2 32.Kf1 Rg5 33.R2h3 Rb2 34.Rf3 Rbb5 35.Rfh3 Rb1+ 36.Ke2 Rb2+ 37.Kf1 Rf5 38.Rh2 Rbb5 39.R4h3 Ra5 40.Kg2 Rae5 41.Kf1 Kh7 42.Rh1 Rg5 43.R1h2 Ref5 44.Rh4 Rd5 45.Ke2 Rge5+ 46.Kf1 Rf5 47.R4h3 Rd2 48.Rg3 Rb2 49.Rgh3 Rb1+ 50.Ke2 Re5+ 51.Kf3 Rb3+ 52.Kg4
83/8- 35:09 62,147,036k 29,463k -1.35 1.Ra1 Re5
83/8- 35:43 63,714,419k 29,721k -1.43 1.Ra1 Re5
83/132- 10:05:47 1,665,633,765k 45,825k -6.13 1.Ra1 Re5

menniepals
Posts: 235
Joined: Wed Mar 08, 2006 7:31 pm
Location: Houston, Texas

Re: 1.g4 opening is losing?

Post by menniepals » Tue Oct 06, 2020 3:57 am

Of course, it is. Only for blitz, it is okay.

mmt
Posts: 343
Joined: Sun Aug 25, 2019 6:33 am
Full name: .

Re: 1.g4 opening is losing?

Post by mmt » Fri Oct 09, 2020 3:27 am

mmt wrote:
Tue Oct 06, 2020 1:40 am
Something unusual: the eval was -1.26 from depth 21 to depth 82 and it took just 34 minutes to get there. But now it's been searching depth 83 for a very long time:
Just to update I gave up on waiting after a couple of days but I will re-evaluate once there is a newer version of SF that behaves differently here.

Post Reply