Page 1 of 1
interesting position from CCC 8
Posted: Fri May 03, 2019 2:32 pm
by jdart
Qxg5 here (from Komodo MC-Xiphos) seems quite difficult for non-NN engines. It is not obvious that giving up the queen is the most powerful move.
[d]2b3r1/1p1p2pk/2nBp2p/2P1Ppq1/R7/4QBPP/5P2/5K2 w - -
Stockfish 10, 4 cores, 30 minutes:
depth 54 seldepth 81 multipv 1 score cp 358 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3g5 h6g5 f1e2 g8h8 e2d3 h7g6 f3g2 b7b5 c5b6 c8b7\
d3d2 g5g4 h3g4 h8h2 g4f5 e6f5 g2c6 b7c6 a4a7 h2f2 d2c3 g6f7 b6b7 c6b7 a7b7 f7e6 b7b8 e6d5 b8b5 d5e4 b5b4 e4d5 b4f4 f2g2 f4f5 g2g3 c3d2 d5e6 f5f8 g3g2 d2e3 e6d5 \
f8d8 d5e6 e3f4 g2g1 d8e8 e6d5 f4f5 g1g2 e8d8 g2f2 f5g4 d5c6 g4g5 f2g2 g5f5 g2g1 d8c8 c6d5 c8c5 d5d4 c5c7 g7g6 f5f6 g1f1 f6g6 f1g1 g6f6
depth 53 seldepth 54 multipv 2 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3c3 g5d8 c3b2 d8e8 h3h4 g7g5 f3e2 e8f7 b2d2 f7g7\
d2e3 g5g4 f1g2 c6e7 e2d3 e7c6 g2h2 g8e8 d3c2 e8g8 e3c3 c6e7 c2d3 e7c6 h2g1 g8h8 c3b2 h8g8 g1g2 c6e7 a4a8 e7c6 a8a1 g8e8 b2c3 h7g6 d3e2 g7f7 a1b1 g6g7 c3e3 c6e7 \
e3d2 e8h8 e2d3 e7d5 g2g1 d5e7 d3c4 e7c6 d2f4 f7g6 g1g2
depth 53 seldepth 54 multipv 3 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3e1 g5d8 e1e3 g8h8 f1g1 d8g5 e3c3 h8g8 g1g2 g5d8\
c3b3 d8e8 h3h4 g7g5 b3e3 g5g4 f3e2 c6e7 e3d2 e7d5 d2b2 e8f7 b2d4 f7g7 e2d3 d5e7 g2g1 g7f7 d3c4 e7c6 d4d2 f7g7 a4a8 g8h8 d2e3 h7g6 e3f4 h8g8 c4d3 g6h7 a8a4 g7f7 \
g1g2 f7g7 g2h2 g7f7 d3f1 f7g7 h2g1 g8e8 f1d3 e8g8 f4e3 g8e8 a4a8
++ solved in 396.578 seconds (3377191500 nodes)
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 3:03 pm
by jp
jdart wrote: ↑Fri May 03, 2019 2:32 pm
Qxg5 here (from Komodo MC-Xiphos) seems quite difficult for non-NN engines. It is not obvious that giving up the queen is the most powerful move.
How can we be sure it is the best move?
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 3:18 pm
by jdart
How can we be sure it is the best move?
I posted multi-PV analysis. But it is always possible a deeper search will show something different.
--Jon
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 3:43 pm
by Joerg Oster
jdart wrote: ↑Fri May 03, 2019 2:32 pm
Qxg5 here (from Komodo MC-Xiphos) seems quite difficult for non-NN engines. It is not obvious that giving up the queen is the most powerful move.
[d]2b3r1/1p1p2pk/2nBp2p/2P1Ppq1/R7/4QBPP/5P2/5K2 w - -
Stockfish 10, 4 cores, 30 minutes:
depth 54 seldepth 81 multipv 1 score cp 358 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3g5 h6g5 f1e2 g8h8 e2d3 h7g6 f3g2 b7b5 c5b6 c8b7\
d3d2 g5g4 h3g4 h8h2 g4f5 e6f5 g2c6 b7c6 a4a7 h2f2 d2c3 g6f7 b6b7 c6b7 a7b7 f7e6 b7b8 e6d5 b8b5 d5e4 b5b4 e4d5 b4f4 f2g2 f4f5 g2g3 c3d2 d5e6 f5f8 g3g2 d2e3 e6d5 \
f8d8 d5e6 e3f4 g2g1 d8e8 e6d5 f4f5 g1g2 e8d8 g2f2 f5g4 d5c6 g4g5 f2g2 g5f5 g2g1 d8c8 c6d5 c8c5 d5d4 c5c7 g7g6 f5f6 g1f1 f6g6 f1g1 g6f6
depth 53 seldepth 54 multipv 2 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3c3 g5d8 c3b2 d8e8 h3h4 g7g5 f3e2 e8f7 b2d2 f7g7\
d2e3 g5g4 f1g2 c6e7 e2d3 e7c6 g2h2 g8e8 d3c2 e8g8 e3c3 c6e7 c2d3 e7c6 h2g1 g8h8 c3b2 h8g8 g1g2 c6e7 a4a8 e7c6 a8a1 g8e8 b2c3 h7g6 d3e2 g7f7 a1b1 g6g7 c3e3 c6e7 \
e3d2 e8h8 e2d3 e7d5 g2g1 d5e7 d3c4 e7c6 d2f4 f7g6 g1g2
depth 53 seldepth 54 multipv 3 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3e1 g5d8 e1e3 g8h8 f1g1 d8g5 e3c3 h8g8 g1g2 g5d8\
c3b3 d8e8 h3h4 g7g5 b3e3 g5g4 f3e2 c6e7 e3d2 e7d5 d2b2 e8f7 b2d4 f7g7 e2d3 d5e7 g2g1 g7f7 d3c4 e7c6 d4d2 f7g7 a4a8 g8h8 d2e3 h7g6 e3f4 h8g8 c4d3 g6h7 a8a4 g7f7 \
g1g2 f7g7 g2h2 g7f7 d3f1 f7g7 h2g1 g8e8 f1d3 e8g8 f4e3 g8e8 a4a8
++ solved in 396.578 seconds (3377191500 nodes)
The MCTS-like scores show Qxg5 clearly superior.
(3 Threads, 1GB Hash, MultiPV=5, 3 minutes search time)
- info depth 39 seldepth 61 multipv 1 score cp 280 mcts score cp 320 visits 74313943 nodes 700028052 nps 3889023 hashfull 1000 tbhits 198927 time 180001 pv e3g5
info depth 38 seldepth 38 multipv 2 score cp 160 mcts score cp 160 visits 29617179 nodes 700028052 nps 3889023 hashfull 1000 tbhits 198927 time 180001 pv f1e1 g5d8 e1f1 d8e8 f1g1 e8f7 a4h4 g7g5 h4a4 h7g7 e3e2 g7h8 f3h5 f7g7 g1h2 h8h7 h2g2 c6e7 h5f3 e7c6 g2g1 c6e7 g1h2 h7h8 e2d2 e7c6 d2c3 h8h7 c3b2 h7h8 f3e2 c6e7 h2g1 e7c6 g1g2 c6e7 e2d3 g7f7
info depth 38 seldepth 41 multipv 3 score cp 160 mcts score cp 156 visits 16148757 nodes 700028052 nps 3889023 hashfull 1000 tbhits 198927 time 180001 pv a4f4 g5d8 f4a4 d8e8 f1g1 e8f7 a4h4 g7g5 h4a4 h7g7 e3e2 g7h8 f3h5 f7g7 g1h2 h8h7 h2g2 c6e7 h5f3 e7c6 g2g1 c6e7 g1h2 h7h8 e2d2 e7c6 d2c3 h8h7 c3b2 g7f7 a4a8 f7g7 h2g2 g7f7 b2c3 h7h8 a8a1 h8h7 a1a4 f7e8 c3b2
info depth 38 seldepth 38 multipv 4 score cp 160 mcts score cp 136 visits 16064402 nodes 700028052 nps 3889023 hashfull 1000 tbhits 198927 time 180001 pv e3e1 g5d8 e1b1 d8e8 b1b2 g7g5 f1g2 e8f7 b2e2 f7g7 g2g1 c6e7 g1h2 h7h8 e2d2 e7c6 d2c3 h8h7 c3b2 g7f7 a4a8 f7g7 h2g2 g7f7 b2c3 h7h8 a8a1 h8h7 g2g1 c6e7 a1a8 e7c6 f3e2 c6e7 c3d2 h7h8 e2f3 e7c6
info depth 38 seldepth 39 multipv 5 score cp 160 mcts score cp 142 visits 36894365 nodes 700028052 nps 3889023 hashfull 1000 tbhits 198927 time 180001 pv f1e2 g5d8 e3b3 d8e8 e2f1 e8f7 b3b2 g7g5 f1g2 h7h8 f3e2 c6e7 b2d2 h8h7 g2g1 f7g7 a4a8 e7c6 d2b2 g7f7 b2c3 h7h8 a8a4 h8g7 g1g2 c6e7 e2f3 g8e8 c3e3 e8g8 a4a8 e7c6 e3c3 g7h8 g2h2 h8h7 h2g1 f7g7
bestmove e3g5 ponder h6g5
Of course, once the move is found A/B-score is also clearly the best.
But this might explain why MCTS-Engines don't have that hard time to find this.
P.S. This mcts-like score is simply the average of all returned scores during the search of each root move.
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 3:48 pm
by jp
jdart wrote: ↑Fri May 03, 2019 3:18 pm
How can we be sure it is the best move?
I posted multi-PV analysis. But it is always possible a deeper search will show something different.
I see. I originally saw "multipv 1" and didn't see the "multipv 2" etc.
I take it you're saying that that depth was the first time SF liked the move.
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 4:02 pm
by Vinvin
jp wrote: ↑Fri May 03, 2019 3:03 pm
jdart wrote: ↑Fri May 03, 2019 2:32 pm
Qxg5 here (from Komodo MC-Xiphos) seems quite difficult for non-NN engines. It is not obvious that giving up the queen is the most powerful move.
How can we be sure it is the best move?
After playing the move, let SF analyse a bit, go back one move and start a "2 best moves" analyse :
Code: Select all
Stockfish_190305_x64_modern:
---------------------------------------------------------------------------
39/40 02:29 964.402k 6.434k +1,50 1.Rf4 Qd8 2.Ra4 Rh8 3.h4 Ne7 4.Kg1 Qe8 5.Qd2 Qf7 6.Kg2 Rg8 7.Qd3 Nc6 8.Qe3 Qe8 9.Kf1 Qd8 10.Qb3 Qe8 11.Qd1 Qd8 12.Qb1 Ne7 13.Bg2 Qe8 14.f4 Qf7 15.Qd3 Nd5 16.Bf3 Ne7 17.Qd2 Nc6 18.Kg2 Qe8 19.Qd3 Ne7 20.Kh2 Qf7
39/65 02:29 964.402k 6.434k +3,13 1.Qxg5 hxg5 2.Ke2 Kg6 3.Kd3 Kf7 4.Ra8 Nb4+ 5.Kc4 Nd5 6.Bxd5 exd5+ 7.Kxd5 b5 8.e6+ dxe6+ 9.Ke5 b4 10.c6 b3 11.Rb8 Ba6 12.Rxb3 Rc8 13.Rb7+ Kg6 14.Rb6 Bf1 15.h4 gxh4 16.gxh4 Kf7 17.c7 g6 18.Rb4 Bd3 19.Rb8 Ba6 20.f3 f4 21.Rb4 Bd3 22.Rxf4+ Bf5 23.Rd4 Bh3 24.Kf4 Kf6 25.Be5+ Ke7 26.Rd2
Re: interesting position from CCC 8
Posted: Fri May 03, 2019 4:12 pm
by camava
jdart wrote: ↑Fri May 03, 2019 2:32 pm
Qxg5 here (from Komodo MC-Xiphos) seems quite difficult for non-NN engines. It is not obvious that giving up the queen is the most powerful move.
[d]2b3r1/1p1p2pk/2nBp2p/2P1Ppq1/R7/4QBPP/5P2/5K2 w - -
Stockfish 10, 4 cores, 30 minutes:
depth 54 seldepth 81 multipv 1 score cp 358 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3g5 h6g5 f1e2 g8h8 e2d3 h7g6 f3g2 b7b5 c5b6 c8b7\
d3d2 g5g4 h3g4 h8h2 g4f5 e6f5 g2c6 b7c6 a4a7 h2f2 d2c3 g6f7 b6b7 c6b7 a7b7 f7e6 b7b8 e6d5 b8b5 d5e4 b5b4 e4d5 b4f4 f2g2 f4f5 g2g3 c3d2 d5e6 f5f8 g3g2 d2e3 e6d5 \
f8d8 d5e6 e3f4 g2g1 d8e8 e6d5 f4f5 g1g2 e8d8 g2f2 f5g4 d5c6 g4g5 f2g2 g5f5 g2g1 d8c8 c6d5 c8c5 d5d4 c5c7 g7g6 f5f6 g1f1 f6g6 f1g1 g6f6
depth 53 seldepth 54 multipv 2 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3c3 g5d8 c3b2 d8e8 h3h4 g7g5 f3e2 e8f7 b2d2 f7g7\
d2e3 g5g4 f1g2 c6e7 e2d3 e7c6 g2h2 g8e8 d3c2 e8g8 e3c3 c6e7 c2d3 e7c6 h2g1 g8h8 c3b2 h8g8 g1g2 c6e7 a4a8 e7c6 a8a1 g8e8 b2c3 h7g6 d3e2 g7f7 a1b1 g6g7 c3e3 c6e7 \
e3d2 e8h8 e2d3 e7d5 g2g1 d5e7 d3c4 e7c6 d2f4 f7g6 g1g2
depth 53 seldepth 54 multipv 3 score cp 167 nodes 16305499409 nps 9058605 hashfull 972 tbhits 0 time 1800001 pv e3e1 g5d8 e1e3 g8h8 f1g1 d8g5 e3c3 h8g8 g1g2 g5d8\
c3b3 d8e8 h3h4 g7g5 b3e3 g5g4 f3e2 c6e7 e3d2 e7d5 d2b2 e8f7 b2d4 f7g7 e2d3 d5e7 g2g1 g7f7 d3c4 e7c6 d4d2 f7g7 a4a8 g8h8 d2e3 h7g6 e3f4 h8g8 c4d3 g6h7 a8a4 g7f7 \
g1g2 f7g7 g2h2 g7f7 d3f1 f7g7 h2g1 g8e8 f1d3 e8g8 f4e3 g8e8 a4a8
++ solved in 396.578 seconds (3377191500 nodes)
Hi mr jdart.
1.e3e1,g5d8 2.e1e3,g8h8(this is what you have)
but if 2.......d8g5. you have initial position,so 2.e1e3,... its not a good move cos blacks can return to initial like whites did