Closed rwbc closed 3 years ago
This bug seems still there in version 0.3.2 (my compilation)
Below is analysis from Megalodon 0.3.2 with feeding the pgn (while analysing) one ply before the illegal move happened:
dep score nodes time (not shown: tbhits knps seldep)
6 -2.70 11.1M 0:36.16 h2h4 g7d4 g1e2 d4b2 h1g1 g8h8
5 -0.05 1.15M 0:03.59 g1e2 d8h4 e2g3 g7d4 f3b7
4 -1.29 65805 0:00.22 g1e2 g8h8 e2f4 g7d4
3 -0.13 10705 0:00.02 e1f1 g7d4 f3b7
2 -1.29 527 0:00.00 g1e2 g7d4
1 +0.84 31 0:00.00 f3b7
I thought this was already fixed? BTW very strange moves with the Q from move 12 'til the illegal end.
[Event "Test@CuteCGdv"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.03"]
[Round "6"]
[White "Megalodon_033r2046-64"]
[Black "Zoe_01"]
[Result "0-1"]
[ECO "B20"]
[GameDuration "00:00:28"]
[GameEndTime "2021-04-03T12:10:02.072 Mitteleuropäische Sommerzeit"]
[GameStartTime "2021-04-03T12:09:33.278 Mitteleuropäische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "28"]
[Termination "illegal move"]
[TimeControl "40/60"]
1. e4 {book} c5 {book} 2. d3 {book} g6 {book} 3. f4 {book} Bg7 {book}
4. Be3 {+0.06/3 0.38s} Bxb2 {+1.85/6 2.96s} 5. Nd2 {-4.56/5 0.37s}
Nf6 {+3.05/5 2.02s} 6. Rb1 {-2.88/5 0.37s} Bd4 {-0.80/6 2.12s}
7. Bxd4 {-1.44/5 0.38s} cxd4 {-1.10/6 2.11s} 8. Ngf3 {+0.32/3 0.38s}
d5 {-1.00/6 3.22s} 9. exd5 {-1.18/5 0.39s} Nxd5 {+0.20/6 3.02s}
10. g3 {+1.24/3 0.38s} Nc3 {+2.00/6 3.58s} 11. Qc1 {-1.68/3 0.40s}
Nxb1 {+6.40/5 1.40s} 12. Qb2 {-3.39/5 0.42s} Nc3 {+4.30/6 1.90s}
13. Qxc3 {-3.18/5 0.41s} dxc3 {+14.70/5 0.54s} 14. Ne5 {-8.52/3 0.41s}
cxd2+ {+18.00/5 1.13s, White makes an illegal move: f1g2} 0-1
I also experienced strange blunders with Megalodon. I think it should be fixed in 6dffa5b46afa2ab257df1cbfabb3e8641f94f7b7, which reduced the imbalance eval weight.
I am still following progress of Megalodon (which isn't very easy due to lots of commits).
My last compilation of version 0.3.0 finally played much better than those before, but I still caught an illegal move already in game 1. (played 6 games vs. Alouette 0.14)
BTW the source for 0.3.0 had the wrong version number in uci id, it still showed 0.2.6.