RelaxChess 7 is available

[Richard1964]

It reacts for sure on command "position startpos moves", depth is not implemented, that because architecture is not built on UCI (and it will not be!) it is only used to play in GUI supported UCI as Arena. RelaxChess is not built to analyze the move all the time but to enjoy the game!. You do not have to go deep if you do not like to swim !. RelaxChess has probably one of the best architecture to jump in and change or built new functionality and You can do it in Java, C++ or C#. The GUI You can develop in Java, C# or Javascript. Which other chess game/engine supports it ?.

[Richard1964]

And besides that UCI protocol is based on pipes, from the technological point of view it is the most primitive way of communication between the 2 programs. And using latest Windows 10 and Universal Windows Platform framework it is standard not possible to use pipes in C# programming language.

[hgm]

Where did you get that nonsense?

[IanKennedy]

Here a quick bullet test vs. Neg 1.3 at that time in CuteChess

What? NEG 1.3d wins games???

Then 1500 for the opponentis really optimistic. NEG is a non-searching engine, and my conclusion was that non-searching engines are unmeasurably weaker than the weakest alpha-beta engines. The fact that rating calculators only place them about 1000 Elo below the weakest AB engines is only because there is nothing in the gap between them to push the ratings further apart. And that some weak AB engines hand out free points even to random movers by forfeiting on time or due to illegal moves before the opponent gets any say in how the game will develop. The bell-curve rating model is just not applicable to engines that can even manage to lose from infinitely weak opponents because of their own bugs.

The true rating of NEG is probably around -4000, on the CCRL scale.

Yes but... NEG isn't a true random mover, it does a sneaky little extra on top. If you play Pos it crushes it.

Engine Score NE Po S-B
1: NEG1_2 10.0/10 ·········· 1111111111 0.00
2: Pos-1.20 0.0/10 0000000000 ·········· 0.00

10 games played / Tournament is finished
Name of the tournament: NEG pos
Level: Blitz 2/1

On my machine its 100% over both Pos and Leela with the random backend.

[hgm]

Indeed. It was actually designed to crush Pos; the name was inspired by that goal. (Although officially NEG is the acronym for "Niet Erg Goed", which is Dutch for "Not Very Good".) I wanted to demonstrate that even without search you could do far better than Pos did. Pos isn't a true random mover either, btw. IIRC it scores about 75% against one, mostly by draws, although it occasionally loses a game. The habit to rush its King towards the center in the early opening is not really helpful...

The current version of NEG plays by counting number of attackers and protectors of each piece, and uses that information, plus the piece value of the lowest attacker and the victim to decide whether it can profitably capture, or (if on a piece of his own) has to move it away. And it is biased towards delivering checks, expected material gains being equal. Especially checks with new pieces (i.e. that did not already check on the previous move). That turns out to be a very effective way of checkmating when you have a significant material lead (where the risk for stalemate furthermore could be large if you did not check). It reduces the risk for stalemate further by refraining from capturing the opponents last minor, if sufficiently much ahead.

Most obvious flaw of this algorithm is that it does not know that threats can often be solved by capturing the attacker. If a Knight delivers a suicidal check to Ke8 from f6 (perhaps because it just captured the Knight there), it would move away his King rather than capture that Knight, because rescueing the King has priority over gaining a Knight, and it doesn't understand that the capture does both. I still want to fix this one day. Even if it would be only for the case where there is a single attacker (where you can be sure it solves the threat without further calculation).

Anyway, it would get totally crushed by a search of 1 ply plus quiescence. So either RelaxChess it totally buggy, or it is a non-searching engine too. I have no idea whether it moves instantly, like NEG does at any TC, or that it uses serious time, in bullet games.

[Guenther]

...
Yes currently the engine always selects Queen as a promotion piece, I still do not understand why You would like to have another piece :-).
...

You did not understand, it promoted to nothing! Promotion in uci is something like e.g. e8=Q not e8, that's what it did in August.

[hgm]

More precisely, e7e8q, not e7e8.

[Guenther]

More precisely, e7e8q, not e7e8.

Left it out, because I am in time trouble. More tomorrow...

BTW I have no idea how strong NEG 1.3 is, but it sure is much stronger than Neg 0.3 which was the only one tested in CCRL.
I wanted to give an upper stimate in fact it was very difficult to find appropriate opponents for Relax in end of August.

Anyhow in my post I concentrated more on the obvious bugs not on strength.

[hgm]

Well, I did a quick test, playing a few 100-game matches between Fairy-Max instances at various fixed depths. The instance that searches 1 ply deeper typically scores between 75% and 80%, i.e. about 200 Elo per ply. The 6-ply instance (which still uses only a few seconds per game) thus is about 1000 Elo stronger than the 1-ply instance. The latter beats NEG 1.3 by about 70%, but is still highly overrating NEG: almost all points it scores are due to it being stalemated by Fairy-Max, in positions where the latter is 3 or 4 Queens ahead. Fairy-Max does not recognize stalemate at 1 ply. When I tried NEG against QueeNy at 1 ply (which does recognize stalemate at 1 ply, but has a grossly mistuned evaluation for the benefit of 3Q vs 7N games), QueeNy beat it 81.5-0.5 (one 50-move draw in KQPPPK). The 1-ply engines are proably about 1500 Elo below TSCP at 1-min/game, but they still seem unmeasurably stronger than NEG 1.3, provided they are able to recognize (and thus avoid) stalemate.

[Guenther]

...
Yes currently the engine always selects Queen as a promotion piece, I still do not understand why You would like to have another piece :-).
...

You did not understand, it promoted to nothing! Promotion in uci is something like e.g. e8=Q not e8, that's what it did in August.

Two examples with log snippet played in CuteChess, Winboard would make no difference...
(clicking on c8 gives the full pgn data)

Code: Select all

325.698: > position startpos moves d2d4 f7f5 c2c4 g8f6 g2g3 e7e6 b1c3 f8b4 a2a3 b4a5 b2b4 a5b6 c4c5 f6d5 c3b1 d5b4 a3b4 c7c6 c5b6 d8b6 c1a3 b6a6 d1c2 g7g6 d4d5 e6d5 a3b2 a6c4 c2c4 d5c4 b2h8 b8a6 e2e4 a6b4 h8c3 b4c2 e1d1 c2a1 c3a1 a7a6 e4f5 b7b5 f5f6 e8d8 a1c3 h7h5 d1d2 g6g5 d2e3 g5g4 e3f4 d7d5 f4g5 b5b4 c3b4 a6a5 b4c5 a8b8 b1c3 b8b2 g5h5 d8c7 f2f3 g4f3 g1f3 b2b3 c5d4 c7b7 f6f7 c8g4 h5g4 b3c3 d4c3 d5d4 c3a5 b7a6 a5b4 c6c5 b4c5 d4d3
325.698: > isready
325.698: < readyok
325.698: > go wtime 199108 btime 236250 movestogo 40
326.338: < bestmove f7f8

[pgn]
[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO"]
[Date "2020.10.09"]
[Round "2"]
[White "Relax_72"]
[Black "Alouette_009-64"]
[Result "0-1"]
[ECO "A86"]
[GameDuration "00:00:45"]
[GameEndTime "2020-10-09T09:40:37.328 Mitteleuropäische Sommerzeit"]
[GameStartTime "2020-10-09T09:39:51.898 Mitteleuropäische Sommerzeit"]
[Opening "Dutch with c4 & g3"]
[PlyCount "80"]
[Termination "illegal move"]
[TimeControl "40/120"]

1. d4 {book} f5 {book} 2. c4 {book} Nf6 {book} 3. g3 {book} e6 {book}
4. Nc3 {2.35s} Bb4 5. a3 {1.85s} Ba5 6. b4 {1.81s} Bb6 7. c5 {1.07s} Nd5
8. Nb1 {1.48s} Nxb4 9. axb4 {0.69s} c6 10. cxb6 {1.09s} Qxb6 11. Ba3 {1.06s} Qa6
12. Qc2 {0.91s} g6 13. d5 {6.91s} exd5 14. Bb2 {1.63s} Qc4 15. Qxc4 {1.09s} dxc4
16. Bxh8 Na6 17. e4 {0.54s} Nxb4 18. Bc3 {0.72s} Nc2+ 19. Kd1 {0.54s} Nxa1
20. Bxa1 a6 21. exf5 {0.56s} b5 22. f6 {0.50s} Kd8 23. Bc3 {0.55s} h5
24. Kd2 {1.46s} g5 25. Ke3 {1.48s} g4 26. Kf4 {1.29s} d5 27. Kg5 {1.31s} b4
28. Bxb4 a5 29. Bc5 {1.33s} Rb8 30. Nc3 {1.34s} Rb2 31. Kxh5 {1.43s} Kc7
32. f3 {0.58s} gxf3 33. Nxf3 {0.58s} Rb3 34. Bd4 {0.60s} Kb7 35. f7 {0.87s} Bg4+
36. Kxg4 Rxc3 37. Bxc3 d4 38. Bxa5 Ka6 39. Bb4 c5 40. Bxc5
d3 {White makes an illegal move: f7f8} 0-1[/pgn]

Code: Select all

465.413: > position startpos moves e2e4 c7c6 d2d4 d7d5 e4e5 c8f5 c2c3 f5b1 a1b1 f7f6 d1h5 g7g6 h5f3 f6e5 d4e5 d8c8 g1e2 c8f5 b1a1 f5e5 c1f4 e5f5 g2g4 f5c2 f4e5 g8f6 b2b3 f8g7 g4g5 c2f5 f3f5 g6f5 g5f6 g7f6 e5d4 e8g8 h1g1 g8f7 d4c5 b7b6 c5e3 c6c5 e1d1 b8c6 f2f4 f7e8 h2h3 e8d7 a2a4 d7c7 a1c1 c5c4 b3b4 a7a5 b4b5 c6d8 c1c2 f6h4 e3d4 d8e6 d4e3 e6c5 c2a2 h4f6 g1h1 c7d7 e3g1 d7c7 h3h4 c7d7 h4h5 d7c7 h5h6 c7d7 f1h3 e7e6 h3f1 d7c7 g1e3 c7d7 f1h3 d7c7 e3c1 c7d7 c1a3 a8a7 h3f1 a7a8 h1h2 f6d8 f1h3 f8f6 h3f5 e6f5 a3c1 d8e7 c1a3 c5a4 a3e7 d7e6 e7f6 a4c5 f6d4 c5d7 a2b2 a5a4 b2c2 a8a5 c2b2 a4a3 b2b1 a3a2 b1c1
465.413: > isready
465.413: < readyok
465.413: > go wtime 231295 btime 110833 movestogo 24
465.803: < bestmove a2a1

[pgn]
[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO"]
[Date "2020.10.09"]
[Round "3"]
[White "Alouette_009-64"]
[Black "Relax_72"]
[Result "1-0"]
[ECO "B12"]
[GameDuration "00:02:18"]
[GameEndTime "2020-10-09T09:42:56.798 Mitteleuropäische Sommerzeit"]
[GameStartTime "2020-10-09T09:40:38.328 Mitteleuropäische Sommerzeit"]
[Opening "Caro-Kann"]
[PlyCount "113"]
[Termination "illegal move"]
[TimeControl "40/120"]
[Variation "Advance Variation"]

1. e4 {book} c6 {book} 2. d4 {book} d5 {book} 3. e5 {book} Bf5 {book} 4. c3
Bxb1 {2.83s} 5. Rxb1 f6 {13.09s} 6. Qh5+ g6 {0.78s} 7. Qf3 fxe5 {2.49s} 8. dxe5
Qc8 {2.15s} 9. Ne2 Qf5 {1.13s} 10. Ra1 Qxe5 {1.31s} 11. Bf4 Qf5 {1.40s} 12. g4
Qc2 {1.80s} 13. Be5 Nf6 {0.80s} 14. b3 Bg7 {12.44s} 15. g5 Qf5 {1.00s} 16. Qxf5
gxf5 {0.87s} 17. gxf6 Bxf6 {0.54s} 18. Bd4 O-O {0.96s} 19. Rg1+ Kf7 {0.55s}
20. Bc5 b6 {1.45s} 21. Be3 c5 {1.15s} 22. Kd1 Nc6 {1.01s} 23. f4 Ke8 {1.39s}
24. h3 Kd7 {0.68s} 25. a4 Kc7 {1.16s} 26. Rc1 c4 {1.81s} 27. b4 a5 {1.85s}
28. b5 Nd8 {0.82s} 29. Rc2 Bh4 {1.10s} 30. Bd4 Ne6 {11.75s} 31. Be3 Nc5 {0.96s}
32. Ra2 Bf6 {1.72s} 33. Rh1 Kd7 {0.94s} 34. Bg1 Kc7 {0.85s} 35. h4 Kd7 {0.96s}
36. h5 Kc7 {0.99s} 37. h6 Kd7 {1.06s} 38. Bh3 e6 {0.69s} 39. Bf1 Kc7 {0.98s}
40. Be3 Kd7 {1.08s} 41. Bh3 Kc7 {0.99s} 42. Bc1 Kd7 {0.98s} 43. Ba3 Ra7 {11.90s}
44. Bf1 Raa8 {11.67s} 45. Rh2 Bd8 {11.86s} 46. Bh3 Rf6 {0.58s} 47. Bxf5
exf5 {0.67s} 48. Bc1 Be7 {1.04s} 49. Ba3 Nxa4 {0.77s} 50. Bxe7 Ke6 51. Bxf6 Nc5
52. Bd4 Nd7 {6.92s} 53. Rb2 a4 {0.71s} 54. Rc2 Ra5 {0.59s} 55. Rb2 a3 {0.59s}
56. Rb1 a2 57. Rc1 {Black makes an illegal move: a2a1} 1-0[/pgn]