An alternative to uho?

[dkappe]

We all know the uho openings by now. Openings so ugly, not even a mother could love them. But they are so imbalanced that the number of draws is low enough to make engine testing possible.

Now TCEC just ran a bonus to test out at what level engines start consistently drawing Stockfish and Dragon (3.1) from the starting position. The opponents were S23’s QL and L4 engines. After 24 games, Stockfish had 7 draws and 17 wins. Dragon 3.1 had 2 draws and 22 wins. Now this is a really small sample size and it is difficult to conclude anything from it, but I’d really like to see a ranking of SF, Dragon and Leela constructed using this method. Could it be that while SF is the king of the uho openings, Dragon excels at cooking up normal openings from the start position? I’d have to imagine leela would also excel at this.

The bonus is still running. I’m watching with some interest.

[Modern Times]

Are there duplicated openings when you let engines play without a book ? If so, then maybe a 2-move book say could be better.

[dkappe]

Are there duplicated openings when you let engines play without a book ? If so, then maybe a 2-move book say could be better.

I don’t know is the answer, but your suggestion sounds good.

[Albert Silver]

Are there duplicated openings when you let engines play without a book ? If so, then maybe a 2-move book say could be better.

The problem with two-move books and other similar opening suites is that you actually get less opening variety than you imagine. You do start with any two moves but then you're dependent on the preferences of the engine. What that means is that openings it would not naturally play will simply never happen. So a King's Indian Main Line will never happen. Many lines in the Ruy Lopez are unlikely to happen. And many other openings. The only way to have an engine play these is to impose them. Now you might ask why this is desirable. The reason is that you have to ask what an engine is for in the first place. And the primary use of an engine for a human player is not in engine against engine testing but rather to research and develop their opening repertoire. And they will therefore be using that engine on those openings. So it makes sense to be sure that an engine is well developed on those openings, doesn't it? It doesn't have to like them, but you at least want to be sure that they've been tested and have experience in them.

This won't make much sense to non-chess players. And I don't mean people who don't know how to play chess but people who don't compete in chess. They will argue, and I understand why, that two moves will provide endless variety without repeating games. But that's not enough. Imagine I give you a battlefield in which you have a large plain, and three clumps of trees. You can arrange that battlefield in tens or hundreds of thousands of ways so that every battle will effectively be different from the other. That's the variety of a two move opening suite. What you won't get, no matter how many ways you rearrange that scenario, are battles in the ocean, on the beaches, with mountains, and more. And that's what you lose.

[Frank Quisinsky]

Hi there,

main reason for my eng-eng matches since FEOBOS.
More or less I am working since 2012 on it.

A test-set with most aggressive openings (balanced FEOBOS positions).

Much more complicated to find out with eng matches "attacker vs. attacker" (lesser quantity on fast lost games).

At the moment I am working on the basic tournament (with my strongest balanced lines).
If ready (basic tourney) I can test the first 50 positions with 21 engines and can compare the results with the basic tourney.

For that reason different things are important:
I will not lost to many time with engines produced a high move average or lost to many games to fast.
To find out 21 engines for the basic tourney isn't easy. After all I found 19 from TOP-50 I can use!

The better the basic tourney, the better is the comparsion "Opening book vs. test-set".
I think I can reduce the quantity of draws with around 9% and the move average with 7 moves.

This all need a while because later ...
Each of the start positions will get an own ranking ...

After the first 50 postions the next 50 position can be generate etc..
So the group of test-engines must be very strong (basic tourney).

Example:
Stockfish vs. Dragon ...
Only a test with my current TOP-50 lines ... 24% of games (50er test-set = 100 games) ended with 1:0 or 0:1 !!!
Unfortunately: Stockfish have since a while no contempt and move average for draws is to high.
That is really very bad for statistics, with a current Stockfish I can nothing do here!

So I am test the Stockfish version from end of 2021 (very strong) vs. Dragon by Komodo 3 without contempt.

Maybe end of the year I am ready to give the first 100 lines with rankings.

A bit spectacular because with such a test set I produced an other problem?

Attacker vs. engines weaker in king safety after openings (Rybka is a good example or Shredder, Zahak, Bit-Genie and many others). This engines lost directly 30-50 of 100 games very fast with an aggressive test-set. With the final result that the produced Elo can't be right. Fun in testing, but in reality the final results with such an aggressive test-set have nothing to do with the reality of engine strengths.

The reason I lost often interest on this work because normally a balanced book (aggressive and passive lines) are much better for each test.

Best
Frank

[Frank Quisinsky]

[pgn][Event "40 Moves in 8 min"]
[Site "fcp_40in8_secret, FQ-1"]
[Date "2022.08.10"]
[Round "63.9"]
[White "Berserk 9 NN"]
[Black "Fire 8.2"]
[Result "1-0"]

1. e4 {book 0s} c5 {book 0s} 2. Nf3 {book 0s} d6 {book 0s}
3. d4 {book 0s} Nf6 {book 0s} 4. Nc3 {book 0s} cxd4 {book
0s} 5. Nxd4 {book 0s} a6 {book 0s} 6. Bg5 {book 0s} e6
{book 0s} 7. f4 {book 0s} Be7 {book 0s} 8. Qf3 {book 0s}
Qc7 {book 0s} 9. O-O-O {book 0s} Nbd7 {book 0s} 10. Bd3
{book 0s} h6 {book 0s} 11. Qh3 {book 0s} O-O {+0.08/24 23s}
12. Bxh6 {book 0s} gxh6 {+0.08/23 0s} 13. g4 {book 0s
(Qxh6)} Nc5 {+1.00/22 47s} 14. Qxh6 {+3.56/27 8s} Bd8
{+1.00/22 0s (Nh7)} 15. Nf5 {+7.62/30 20s (Rhg1)} Ne8
{+3.57/23 22s} 16. g5 {+8.26/23 12s} d5 {+4.71/26 34s (a5)}
17. Rhg1 {+10.30/23 9s (g6)} Qxf4+ {+10.90/20 20s} 18. Kb1
{+12.08/22 8s} b6 {+15.45/23 41s (Ra7)} 19. Ne2 {+23.98/23
9s} Qe3 {+15.45/22 0s (Qh4)} 20. Nxe3 {+30.46/26 19s} Ra7
{+M12/23 29s (Ng7)} 21. Ng4 {+M13/32 8s} f5 {+M11/29 10s}
22. gxf6 {+M11/39 9s} Bxf6 {+M9/27 1s} 23. Nxf6+ {+M10/44
11s (Rdf1)} Kf7 {+M9/11 0s} 24. Qh7+ {+M9/48 11s} Kxf6
{+M8/24 0s} 25. Qh4+ {+M8/53 9s} Ke5 {+M7/22 0s} 26. Qg3+
{+M7/60 10s} Kf6 {+M6/20 0s} 27. Qg5+ {+M6/72 11s} Kf7
{+M5/11 0s} 28. Rdf1+ {+M5/101 11s (Rgf1+)} Nf6 {+M4/11 0s}
29. Qxf6+ {+M4/124 11s} Ke8 {+M3/11 0s} 30. Qxf8+ {+M3/127
0s} Kd7 {+M2/11 0s} 31. Rg7+ {+M2/127 0s (Rf7+)} Kc6
{+M1/11 0s} 32. Nd4# {+M1/127 0s} 1-0[/pgn]

Interesting position after move 10!
In that example move 11 is to much.
Different engines like to play 11. ... 0-0

But after move 10 many interesting moves are possible.
Also black can win with the right defender moves at first with an attack on queens site after.

Such balanced positions I collect for an test-set.
And used Attacker vs. Attacker for check that (Fire 8.2 is strong in attacking chess).
In this case Berserk 9 is the maestro!



Best
Frank

All isn't easy ... if I collect only such positions ... Wasp vs. Rybka is 99.0 : 1,0 (means attacker vs. an engine have problems with king safety). So, I have to find out a more silent way ... move 9 or 10 and not move 11 as end position for the test-set ... much other lines after move 10 are possible. Only as a small example!

[Modern Times]

<snip> you have to ask what an engine is for in the first place. And the primary use of an engine for a human player is not in engine against engine testing but rather to research and develop their opening repertoire. <snip>

A fact often forgotten on this forum, although I'd say the primary use is surely to analyse their whole games, not just opening repertoire.

Yes there are a minority who like to run engines on playchess, and there is also the competitive element between engine authors, but for most people they are an analysis tool.

[Lazy_Frank]

The problem with two-move books and other similar opening suites is that you actually get less opening variety than you imagine. You do start with any two moves but then you're dependent on the preferences of the engine...

Here is total number of chess positions than arises from standard chess initial position:

depth pos_count
0 1
1 20
2 400
3 5362
4 72078
5 822518
6 9417681
7 96400068
8 988187354
9 9183421888

At depth 4 total number is 72078, i have made filtering there, still 550 positions looks very interesting and unclear.

[peter]

The problem with two-move books and other similar opening suites is that you actually get less opening variety than you imagine. You do start with any two moves but then you're dependent on the preferences of the engine...

Here is total number of chess positions than arises from standard chess initial position:

depth pos_count
0 1
1 20
2 400
3 5362
4 72078
5 822518
6 9417681
7 96400068
8 988187354
9 9183421888

At depth 4 total number is 72078, i have made filtering there, still 550 positions looks very interesting and unclear.

I too think there'd be enough variety with 4 plies already, engines could yet show their even mor realistic "whole- game-strength" in playing the opening on their own too and even positions like 1.d4 Nf6 2.c4 e6 are less balanced (as for draw- death of eng-eng-matches) than main lines of by humans much played and well analysed postions of e.g. 8 full moves.
So one can take intentionally suboptimal moves to get 8-movers unbalanced again, or have the natural advantage of White's at the earlier positions before Black has counterbalanced them by own best moves already given, I'd prefer the latter too,
regards

[dkappe]

The competition has picked up a bit and Minic was able to hold Dragon to two draws. I did just notice that Zahak was able to rope-a-dope Stockfish to two draws. Very nice.

At the moment, leela is climbing through the foothills. I expect it will do even better than Dragon and the trailing Stockfish.

Very interesting.