Balanced opening lines or advantageous opening lines?

[smatovic]

Question, how do you define balanced opening, unbalanced opening, advantageous opening in centi pawns?

--
Srdja

[jefk]

for some odd reason my reaction came into another thread (future of computer chess)

So again:

with sufficiently long time controls, balanced openings will only lead to draws for the strongest
engines. Even unbalanced (but not biased) openings will lead more and more to draws.

That's why in the higher leagues of TCEC some 'advantageous' (biased) lines are chosen;
then ofcourse a specific engine should play such line with both colors.

https://wiki.chessdom.org/Openings_FAQ

[Uri Blass]

the Pohl openings:
https://www.sp-cc.de/uho_2024.htm

these UHO openings look good to me, whererby the drawkiller
openings may be a bit too artificial (to resemble 'real' chess)

That was the main reason to build UHO. Because it is not artificial like my other opening-concepts. There are more than Drawkiller:
https://www.sp-cc.de/anti-draw-openings.htm

In UHO, only lines out of human games from the Megabase are used. Nothing is changed, edited.
And because, I made not only one UHO-opening set, but many, with increasing advantage for white, the draw-ratio in any engine-tournament or test or ratinglist can be adjusted easily to a vlaid range of 45%-60%, by choosing the UHO-eval-range, which works best on your settings:
Eval [+0.85;+0.94] 6mvs: 6696 lines 8mvs: 19303 lines 8mvs_big: 37622 lines
Eval [+0.90;+0.99] 6mvs: 6292 lines 8mvs: 17145 lines 8mvs_big: 33559 lines
Eval [+0.95;+1.04] 6mvs: 5238 lines 8mvs: 14256 lines 8mvs_big: 30372 lines
Eval [+1.00;+1.09] 6mvs: 4740 lines 8mvs: 13227 lines 8mvs_big: 25916 lines
Eval [+1.05;+1.14] 6mvs: 4125 lines 8mvs: 11660 lines 8mvs_big: 22828 lines
Eval [+1.10;+1.19] 6mvs: 3460 lines 8mvs: 9601 lines 8mvs_big: 20600 lines
Eval [+1.15;+1.24] 6mvs: 3161 lines 8mvs: 8940 lines 8mvs_big: 17470 lines
Eval [+1.20;+1.29] 6mvs: 2717 lines 8mvs: 7869 lines 8mvs_big: 15464 lines

https://www.sp-cc.de/uho_2024.htm

And look here, to see, how much better the statisitcs are, when UHO is used, compared to 2 balanced opening-sets:
https://github.com/official-stockfish/S ... -911982470
I think it is fair to say, that UHO blew the balanced openings out of the water...
(3 testruns in Fishtest. 60000 games each, so nobody can doubt these results)

Thats because Vondele (Stockfish maintainer) wrote this later in the linked thread:
"We have quite successfully adopted UHO as the training book. We did find a recent case where there is quite a difference between the Elo measured on the 8moves and the UHO book (#3937) but overall the experience is good. Several major tournaments have adopted UHO-style books as well."

To make this clear: Because Fishtest uses UHO-openings, OpenBench uses UHO-openings, chesscom-tournament use UHO-openings:
At least 99% (more likely 99.9%) of all engine-games, played on this planet, are played, using UHO-openings, right now. (Fishtest alone plays 1-2 billions of games a year using an UHO-book (they build their own huge book from Lichess games, but the concept is still my UHO-concept, of course)). So, today, balanced openings are already dead, considering their amount of using in the worldwide computerchess, compared to UHO...

Maybe more than 99% of the games that you can download use UHO but
I do not believe that you are correct that at least 99% of the engine games are played with UHO

I play engine games without UHO in my computer(I have so far 23750 games for my rating list of fixed nodes per move) and also played games not for my rating list without UHO.
SSDF play without UHO
people who play games in chess sites for rating play without UHO.
CCRL played millions of games without UHO.

It may be interesting to have number for UHO games that you can download and non UHO games that you can download to see the percentage.

Starting with non UHO games

CCRL has 1851872 games at 40/15,1198884 blitz games ,653397 FRC games,2165313 blitz archive games
CEGT has 1797548 games at 40/20,3787230 blitz games 40/4 ,34140 40/120 games,253230 ponder on 3+1 games,517350 ponder on 5+3 games ,82100 25+8 ponder on games
SSDF has 164540 games.

I am not sure how many games lichess has between bots that are not UHO games.

[Peter Berger]

People (especially developers, who might feel bad about this) have to realize how few people care about the general progress of engines the way it is happening right now.
Chess is a drawish game, engines are mighty strong, so it is easy to predict that at some point (probably very soon) we will get UHO openings like 1. e4 h5 2. d4 g6.
There can't be much more than 100 people on this planet who care about such a game chess-wise.
On the other hand, there is the practical question how to measure anything, if you can't tell Rebel 16.3 and current Stockfish development version apart, if you give Rebel its own book and pit them against each other at tournament time control. You'll get mostly draws, obviously.
At the same time chess itself is booming, people do care about the game and its inner truth very much - just look at the interest the Candidates raised.
So, what do people want from chess engines? One thing is: it can be a judge. If it gives 0.00, it is a draw, if it shows +4.62, this is a win - there are certainly several millions of people who use engines this way on a regular basis. Engines have come close to perfection performing this task by now, it is not so interesting if they get any better at this IMHO.
Chess-wise people only care about games that start with equal opening positions, as this is somehow relevant for themselves.
I'd argue ( and yes, I know I am repeating myself ad nauseam) that current testing schemes provide very little to create progress "chess-wise" (as defined above), to the contrary.
I'd like to suggest a different kind of rating list that feels way more relevant to me for human spectators, although I am not ready to invest all the time to create it myself.
Let's say we want to test top engines (within a range of like 100 ELO points) at longer time controls to find out which one is the most useful.
We don't do UHO openings, engines shall play whatever they consider the very best (maybe with some kind of little book for sake of variety).
As we know that they'll create mostly useless draws when pitted against each other, we change the opponent pool instead.
Setting this up in a reasonable way would be the (probably +very+) time-consuming part. We have to find the rating difference that is optimal to create a scenario like the one we get with the UHO openings math-wise. So top engines would play some kind of matches against a range of weaker opponents - I can't scale this without doing all the work, maybe against opponents of like Rybka strength?
The number one in this list (which may still be Stockfish) would be most useful to know for humans to come up with fresh opening ideas in GM games or in correspondence IMHO.

[Uri Blass]

I can add that I do not believe openbench is using only UHO and I think it is the choice of the tester if to use UHO or not UHO.

[Uri Blass]

People (especially developers, who might feel bad about this) have to realize how few people care about the general progress of engines the way it is happening right now.
Chess is a drawish game, engines are mighty strong, so it is easy to predict that at some point (probably very soon) we will get UHO openings like 1. e4 h5 2. d4 g6.
There can't be much more than 100 people on this planet who care about such a game chess-wise.
On the other hand, there is the practical question how to measure anything, if you can't tell Rebel 16.3 and current Stockfish development version apart, if you give Rebel its own book and pit them against each other at tournament time control. You'll get mostly draws, obviously.
At the same time chess itself is booming, people do care about the game and its inner truth very much - just look at the interest the Candidates raised.
So, what do people want from chess engines? One thing is: it can be a judge. If it gives 0.00, it is a draw, if it shows +4.62, this is a win - there are certainly several millions of people who use engines this way on a regular basis. Engines have come close to perfection performing this task by now, it is not so interesting if they get any better at this IMHO.
Chess-wise people only care about games that start with equal opening positions, as this is somehow relevant for themselves.
I'd argue ( and yes, I know I am repeating myself ad nauseam) that current testing schemes provide very little to create progress "chess-wise" (as defined above), to the contrary.
I'd like to suggest a different kind of rating list that feels way more relevant to me for human spectators, although I am not ready to invest all the time to create it myself.
Let's say we want to test top engines (within a range of like 100 ELO points) at longer time controls to find out which one is the most useful.
We don't do UHO openings, engines shall play whatever they consider the very best (maybe with some kind of little book for sake of variety).
As we know that they'll create mostly useless draws when pitted against each other, we change the opponent pool instead.
Setting this up in a reasonable way would be the (probably +very+) time-consuming part. We have to find the rating difference that is optimal to create a scenario like the one we get with the UHO openings math-wise. So top engines would play some kind of matches against a range of weaker opponents - I can't scale this without doing all the work, maybe against opponents of like Rybka strength?
The number one in this list (which may still be Stockfish) would be most useful to know for humans to come up with fresh opening ideas in GM games or in correspondence IMHO.

I agree that it may be interesting to see how engines perform without biased opening book and even if they have 100% draws between themselves there can be difference against weaker engines or against the same engine with relatively small number of nodes.

I think these games may be more interesting for chess players and I am not sure if stockfish will be the best or not.

I think that games of my nobook tournament that at this point of time are only in my computer may be interesting and the same for games of tournaments from some balanced common book positions.

[Graham Banks]

Some interesting opinions and arguments.

[Antihelion]

It depends on the goals.
Balanced openings are the most fair, and the most boring.
Advantageous lines are the least fair, and the most exciting.

I think TCEC has it right for a watched contest. Play polar openings, but force both engines to play as white and black to even it out.
For an unsupervised contest, I don't think it matters.

How is it unfair? I thought such openings are always played in pairs.

[Dann Corbit]

It depends on the goals.
Balanced openings are the most fair, and the most boring.
Advantageous lines are the least fair, and the most exciting.

I think TCEC has it right for a watched contest. Play polar openings, but force both engines to play as white and black to even it out.
For an unsupervised contest, I don't think it matters.

How is it unfair? I thought such openings are always played in pairs.

An engine far below the strength of top five engines may win a game that it would not win against the much stronger engine because the opening is either lost or nearly lost. With equal books a vastly inferior engine would gather less points, inflating its strength by being handed victories and draws it could not obtain on its own.

Suppose that in a ten game match there are two wins handed to the weaker engine.
With equal books, it would probably be something like 4 wins for the stronger side and 6 draws
With the freebies it will be perhaps 5 wins two losses and three draws (recall that the stronger side will not be able to win the two freebies,)

If there are a lot of games, and very few busted openings, it won't matter much. And it is unlikely to change the resolution of the top engines. But it does not give a realistic performance Elo

[Chessqueen]

There is a place for both but I prefer balanced because there is already too many non balanced games.
It is obvious after lc0 beat berserk with 1.g4 in TCEC that balanced opening lines or even no opening book is not always a draw even between top engines.

I test now stockfish15 at x nodes per move against berserk13 at 2x nodes per move(with contempt 0,10,20,30,40,50 when I started with x=1000 and double x after every match.

it is still not obvious if contempt of berserk change the playing strenght of it and it is still not clear if the berserk team is better or worse than stockfish15 in these conditions.

a win with black from the tournament.
White used 1024000 nodes per move.

I also I prefer balanced Opening. Here is another easy win for KomodoDragon 3,3 when in the first 8 moves White was already at a great dvantage

[pgn][Event "TCEC Season 26 - Division P"]
[Site "https://tcec-chess.com"]
[Date "2024.05.02"]
[Round "31.3"]
[White "KomodoDragon 3.3"]
[Black "Berserk 13"]
[Result "1-0"]
[BlackElo "3540"]
[ECO "B00"]
[Opening "Owen Defence"]
[Variation "3.Bd3 e6 4.Nf3"]
[WhiteElo "3592"]
[TimeControl "3600+6"]
[Termination "adjudication"]
[PlyCount "163"]

1. e4 b6 2. d4 Bb7 3. Bd3 e6 4. Nf3 d5 5. exd5 Qxd5 6. c4 Qd8 7. Nc3 Nf6 8.
Qe2 Nbd7 9. Bc2 Bd6 10. Be3 O-O 11. O-O-O c5 12. Rhg1 cxd4 13. Nxd4 Qe7 14.
g4 Rfd8 15. Kb1 Rac8 16. g5 Ne8 17. Bc1 g6 18. h4 Bc5 19. a3 Nb8 20. Nb3
Nc6 21. Rge1 Ba8 22. Nxc5 bxc5 23. Rxd8 Rxd8 24. Qe3 Nd4 25. Be4 Bb7 26.
Bxb7 Qxb7 27. Ne4 a5 28. Qc3 Qc7 29. Qh3 Qb6 30. Bf4 a4 31. Qe3 Qa6 32. Be5
Qxc4 33. Rc1 Qb3 34. Qxb3 axb3 35. Rd1 Rd5 36. f4 h5 37. Nc3 Rd8 38. Na4
Rd5 39. Nb6 Rd8 40. a4 Nd6 41. Bf6 Re8 42. Nd7 N4f5 43. Nxc5 Ra8 44. Rd3
Nc4 45. Rc3 Nfd6 46. Kc1 Na5 47. Nxb3 Nab7 48. Nc5 Nxc5 49. Rxc5 Rxa4 50.
Be5 Ra6 51. Kc2 Kf8 52. b4 Rb6 53. Kb3 Ke7 54. Rc7+ Ke8 55. Bxd6 Rxd6 56.
b5 Rd1 57. Kb4 Rf1 58. Rc4 Ke7 59. b6 Rb1+ 60. Ka5 Kd6 61. Ka6 Kd5 62. Rc7
Ke4 63. Rxf7 Kf3 64. Rf6 Ra1+ 65. Kb7 Rd1 66. Rxe6 Kg4 67. Kb8 Kf5 68. Rf6+
Kg4 69. b7 Re1 70. Ka7 Re7 71. Ka8 Kxh4 72. b8=Q Re3 73. Rxg6 Ra3+ 74. Ra6
Rxa6+ 75. Kb7 Ra2 76. Qe5 Rg2 77. f5 Kh3 78. f6 h4 79. g6 Rg3 80. g7 Rg4
81. Qe6 Kg3 82. Qxg4+ 1-0[/pgn]