Tcheran - UCI chess engine Author: Jonathan Gilchrist
Rating Chess Engines Diary CEDR=2242
If you want to know what the ranking of the Tcheran engine is, we encourage you to visit the website with the monthly updated CEDR ranking - RATING page.
Tcheran 8.0 what's new?
Network #7, upping the size to 384 hidden layer nodes with a fresh dataset generated using network #6 (62.36 +- 16.50)
Improve null move pruning reduction formula (29.37 +- 10.50)
Network #8, upping the size to 512 hidden layer nodes and trained with a larger 1B (unfiltered) FEN dataset (27.07 +- 9.95)
Network #9, upping the size to 768 hidden layer nodes and trained with an even larger 1B (after filtering) dataset and a WDL proportion of 0.4 (27.05 +- 9.85)
Add SEE pruning (25.45 +- 9.40)
Use history gravity and maluses (15.92 +- 7.40)
Do SEE when picking moves instead of when scoring moves (11.32 +- 6.27)
Compute checkers in make_move (8.03 +- 5.01)
Overwrite TT entries from old searches with static eval cache entries (7.85 +- 4.83)
Cache static eval of positions in the transposition table (5.05 +- 3.69)
Re-search with zero-window but full depth after raising alpha in a reduced-depth PVS search (2.83 +- 2.22)
Misc
Update the minimum Rust version to 1.91
Remove paste dev dependency
Remove arrayvec dependency
Simplify the move picker implementation
Simplify move scoring by removing sentinel values
Started tracking the bench in a '.bench' file which can be updated via 'just bench'
Fixed a datagen bug which caused non-decisive games to always be stored with an outcome of a loss for white
Switched to using 'tacticals' terminology for non-quiet moves
Improved 'd eval' command by showing piece contributions similar to Stockfish
Implement 'go nodes' command
Start adding a hash suffix to network files
Derive Copy for ZobristHash
Added a Rust implementation of OpenBench's SPRT algorithm
Bench depth was upped from 10 to 12
Author test:
Elo | 168.06 +- 8.33 (95%)
Conf | 8.0+0.08s Threads=1 Hash=16MB
Games | N: 4096 W: 2310 L: 470 D: 1316
Penta | [14, 104, 501, 886, 543]
Elo | 182.37 +- 14.48 (95%)
Conf | 40.0+0.40s Threads=1 Hash=128MB
Games | N: 1026 W: 572 L: 78 D: 376
Penta | [0, 10, 114, 274, 115]
Tcheran 5.1 vs other engines:
| Fortuna 1.0 JA | 4/4 | +4 | 4 Partii |
| Aurora 20241007 | 2.5/4 | +1 | 4 Partii |
| JudaS ++ 5.0 | 1/4 | -2 | 4 Partii |
| ShashChess 38 | 0/4 | -4 | 4 Partii |
| Alexander 4.0 | 0/4 | -4 | 4 Partii |
| DarkSeid 5.2 | 0/4 | -4 | 4 Partii |
| Artemis 6TR | 0/4 | -4 | 4 Partii |
| Lambergar 1.0 | 0/4 | -4 | 4 Partii |
| Halcyon 1 | 2/3 | +1 | 3 Partii |
| Chessplusplus 1.2.0 | 2/3 | +1 | 3 Partii |
| Belette 3.0.0 | 0.5/3 | -2 | 3 Partii |
| Zeta-Dva 0402 JA | 2/2 | +2 | 2 Partii |
| Albatros 1.0.0 | 2/2 | +2 | 2 Partii |
| MadChess 3.2.2 | 2/2 | +2 | 2 Partii |
| Zagreus 5.0 | 2/2 | +2 | 2 Partii |
| Dinora 0.3.0 | 2/2 | +2 | 2 Partii |
| Omelette 1.0 JA | 2/2 | +2 | 2 Partii |
| Camel 1.6.0 | 2/2 | +2 | 2 Partii |
| StoneChess 2024-09-13 | 2/2 | +2 | 2 Partii |
| Amy 0.9.6-pre AB JA | 2/2 | +2 | 2 Partii |
| Ssehc 1.0 | 2/2 | +2 | 2 Partii |
Tcheran engine top results:
Sample batch of this engine:
[Event "2024.09.14.NewEnginesTest"]
[Site "Chess Engines Diary"]
[Date "2024.09.17"]
[Round "1.32"]
[White "Tcheran 3.0"]
[Black "StoneChess 2024-09-13"]
[Result "1-0"]
[ECO "B00"]
[PlyCount "72"]
[EventDate "2024.??.??"]
{[%evp
0,72,19,31,42,25,25,33,56,29,66,76,56,24,36,21,11,4,-5,-3,16,4,10,-56,-5,-36,-18,-61,-61,-18,-15,-1,7,3,-6,-10,5,-13,9,3,10,13,10,49,46,43,67,31,56,45,49,47,71,43,90,58,61,73,114,133,184,186,153,187,235,243,654,652,873,1006,2044,2497,29979,29984,29985]}
1. e4 Nc6 2. d4 d5 3. e5 f6 4. Nf3 g5 5. h3 Bg7 6. Bb5 Nh6 7. O-O O-O
8. Re1 g4 9. hxg4 Bxg4 10. Bxc6 bxc6 11. e6 Qd6 12. Qe2 Nf5 13. c3 c5
14. Qc2 Bxf3 15. dxc5 Qc6 16. Qxf5 Be4 17. Qh3 Qxc5 18. Be3 Qb5 19. Nd2
Bg6 20. b3 Qd3 21. Nf3 c6 22. Rac1 Qa6 23. a4 Rfb8 24. Nd4 Rb7 25. b4
Rc8 26. a5 c5 27. bxc5 Rxc5 28. Nb3 Rcb5 29. Nc5 Qxa5 30. Nxb7 Rxb7 31.
Red1 Rc7 32. Qf3 Be4 33. Qxe4 Rxc3 34. Qb1 Rxc1 35. Qb8+ Bf8 36. Rxc1
h5 1-0

Comments
Post a Comment