Chessqueen wrote: ↑Sat Feb 06, 2021 3:57 pm
The final score after 10 games was 5 win 2 Loss 3 Draw in favor of Sargon Depth 6, But in the last game Sargon was clearly winning again and it repeated the 3 fold repetition WHY, I thought that it would avoid this in a clearly winning position, Why Sargon did NOT take the g7 pawn ?
[pgn][Event "Computer chess game"]
[Site "DESKTOP-OFQ3C0P"]
[Date "2021.02.06"]
[Round "10"]
[White "Sargon-engine Depth6"]
[Black "Mizar_30"]
[Result "1/2-1/2"]
[BlackElo "1683"]
[ECO "D32"]
[Opening "QGD Tarrasch"]
[Time "08:00:25"]
[Variation "5.Nf3 Nc6 6.e3 Nf6 7.Bb5"]
[WhiteElo "????"]
[TimeControl "2400+5"]
[Termination "normal"]
[PlyCount "89"]
[WhiteType "program"]
[BlackType "program"]
1. Nf3 c5 2. c4 Nf6 3. Nc3 e6 4. e3 Nc6 5. d4 d5 6. cxd5 exd5 7. Bb5 Bd6 8.
dxc5 Bxc5 9. O-O O-O 10. Bd2 Ne4 11. Rc1 Nxd2 12. Qxd2 d4 13. Ne4 Bb6 14.
Bxc6 bxc6 15. Nxd4 Qd5 16. Nc3 Qc4 17. b3 Qa6 18. Nxc6 Qa3 19. Nb5 Qa6 20.
a4 Qb7 21. Qd6 Re8 22. Qd1 Qd7 23. a5 Qxd1 24. Rfxd1 Bg4 25. axb6 Bxd1 26.
Rxd1 axb6 27. Nc7 Kh8 28. Nxe8 Rxe8 29. Rd7 Ra8 30. Kf1 f6 31. Ke2 b5 32.
Na7 b4 33. Nc6 Ra2+ 34. Kf3 h6 35. Rb7 Kh7 36. Nxb4 Ra5 37. Nd3 Rh5 38. Nf4
Re5 39. Rd7 Rg5 40. Ne6 Rf5+ 41. Ke2 Kh8 42. Rd8+ Kh7 43. Rd7 Kh8 44. Rd8+
Kh7 45. Rd7 {3-fold repetition} 1/2-1/2[/pgn]
I tried to reproduce this problem and failed. Stripping back the problem to the essentials, this is how the Sargon V1.01 operates at the UCI command level (this is using the Windows command prompt to run the engine);
uci
id name Sargon 1978 V1.01
id author Dan and Kathe Spracklin, Windows port by Bill Forster
option name FixedDepth type spin min 0 max 20 default 0
option name LogFileName type string default
uciok
position fen 8/3R2pk/4Np1p/5r2/8/1P2PK2/5PPP/8 w - - 9 41 moves f3e2 h7h8 d7d8 h8h7
go depth 6
info depth 1 score cp 625 time 3 nodes 35 nps 11000 pv d8d7
info depth 2 score cp 625 time 18 nodes 293 nps 16000 pv d8d7 f5e5
info depth 3 score cp 625 time 104 nodes 1952 nps 18000 pv d8d7 f5e5 d7g7 h7h8
info depth 6 score cp 625 time 6438 nodes 126490 nps 19000 pv d8d7 f5e5 d7g7 h7h8 g7e7 e5b5
info depth 6 score cp 575 time 13278 nodes 253669 nps 19000 pv g2g4 f5b5 d8d7 b5b3 d7g7 h7h8
bestmove g2g4
The bold lines are the user commands and the rest is Sargon's responses. So the position command sets the position after 40. Rf5+ and then it adds the four following moves 41.Ke2 Kh8 42.Rd8+ Kh7 in UCI notation. It's important that the engine is provided history like this, and not just the current position, otherwise it won't be able to know if there's a repetition.
So now it is Sargon's move and the command go depth 6 asks Sargon to calculate the best move at depth 6. Sargon V1.00 would calculate lower levels and report on them. Sargon V1.01 saves time by only calculating lower levels 1,2 and 3 (which are essentially instantaneous), in case there is an immediate mate, then skipping all levels until the target depth.
Look at these two lines generated by Sargon;
info depth 6 score cp 625 time 6438 nodes 126490 nps 19000 pv d8d7 f5e5 d7g7 h7h8 g7e7 e5b5
info depth 6 score cp 575 time 13278 nodes 253669 nps 19000 pv g2g4 f5b5 d8d7 b5b3 d7g7 h7h8
In the first one we see that Sargon wants to play d8d7, that is 43. Rd7 at depth 6. It's a pity because at depths 1,2 and 3 Sargon would play Rxg7 (you can see that's the first move of its PV in the depth 1,2 and 3 reports). Sargon sees that Black cannot prevent Rxg7 (the move shows up later in the PV). For some reason sometimes it is like a cat that prefers to play with the mouse rather than kill it right away. It was a common problem with old computer programs. The most famous example is Coko v Genie 1971, see
https://www.chessprogramming.org/Coko.
My main improvement in V1.01 compared to V1.00 is that instead of accepting Sargon's decision to play 43. Rd7 repeating the position, I reject that move and ask Sargon to try again (internally, in the UCI wrapper around the old Sargon code). Unfortunately it still doesn't play Rxg7, it pushes the terrified mouse the other way with 43. g4 instead. Again Sargon is content that it can take the g pawn later (it's in the PV starting with g2g4 too). Sargon reports that the best move is 43. g4.
Chessqueen wrote: ↑Sat Feb 06, 2021 9:51 pm
Bill there is another problem that i detected in sargon...
It's pretty obviously the same problem.
Chessqueen wrote: ↑Sun Feb 07, 2021 3:48 am
Yes I downloaded the last version that you fixed and To set up sargon to play at depth6 under Arena Gui, I click under Engines then select engine1 or engine2 and to the right ===> where it say configure I select the fixed depth to 6 and save it as a new engine and give it a new name such as sargon-engine Depth6, then I go to Load engine and select sargon-engine Depth6
Now I've seen this I had to repeat and alter my test procedure a little. This time the depth is specified with my FixedDepth option, not by a depth parameter to the go command. It's similar to the above test, and with the same result;
uci
id name Sargon 1978 V1.01
id author Dan and Kathe Spracklin, Windows port by Bill Forster
option name FixedDepth type spin min 0 max 20 default 0
option name LogFileName type string default
uciok
setoption name FixedDepth value 6
position fen 8/3R2pk/4Np1p/5r2/8/1P2PK2/5PPP/8 w - - 9 41 moves f3e2 h7h8 d7d8 h8h7
go
info depth 1 score cp 625 time 3 nodes 35 nps 11000 pv d8d7
info depth 2 score cp 625 time 21 nodes 293 nps 13000 pv d8d7 f5e5
info depth 3 score cp 625 time 121 nodes 1952 nps 16000 pv d8d7 f5e5 d7g7 h7h8
info depth 6 score cp 625 time 7876 nodes 126490 nps 16000 pv d8d7 f5e5 d7g7 h7h8 g7e7 e5b5
info depth 6 score cp 575 time 15845 nodes 253669 nps 16000 pv g2g4 f5b5 d8d7 b5b3 d7g7 h7h8
bestmove g2g4
I am convinced you are not running the right version of the engine. I opened up Arena for the first time since I was using V1.00 and the procedure for changing to V1.01 is confusing at best. Unfortunately Arena reports the engine author rather than the engine name. The engine author hasn't changed (the engine name includes the version number).
The easiest way to tell the difference is in the PV reports that Arena shows while Sargon is thinking. In V1.01 they will go 1,2,3,6 as shown above. In V1.00 they would either skip lower levels and go to 6 directly, or they would skip 1 and 2 only, then go 3,4,5,6. (It depends on how you specified fixed depth).
The next easiest way to tell the difference is to configure a LogFileName for Sargon within Arena and to look at the contents of that file after a run. For V1.01 there will be many mentions of a "Loop state machine" but that text won't appear in V1.00 log files because there was no loop state machine in that version.
If you are convinced that you have correctly installed Sargon V1.01 and it is really repeating in trivial ways like this, please configure the LogFileName and send me the log next time you observe the problem. This kind of problem analysis is time consuming (I just spent most of my Sunday afternoon) and some detailed information would make a big difference.