Giving up the center

[Karol Majewski]

I have a question to a similar position:

[d]r1bqkb1r/ppp2ppp/2np1n2/8/3NP3/2N5/PPP2PPP/R1BQKB1R w KQkq - 0 6

This can happen out of Philidor or Scotch, so that's very common position on an amateur level. So what's White's best way to continue? I guess many good moves here. Leela favours Bb5, while Stockfish switches between Bf4, Be2, Nxc6. Can someone with a strong CPU run Stockfish for a long time in this position?

As you say, Stockfish's choices vary considerably from iteration to iteration. Here's what Stockfish-dev offers at depth 50:

info depth 50 seldepth 83 multipv 1 score cp 84 nodes 95191474646 nps 29182257 hashfull 1000 tbhits 0 time 3261964 pv d4c6 b7c6 f1d3 f8e7 e1g1 e8g8 f2f4 a7a5 h2h3 c8b7 c1d2 f6d7 d1f3 e7f6 a1d1 f6d4 d2e3 d4e3 f3e3 d7c5 d3c4 a5a4 f1e1 d8b8 e4e5 b7c8 e3d4 c8e6 b2b3 a4b3 a2b3 e6c4 d4c4 b8b6 g1h1 d6d5 c4d4 c5d7 d4d2 f8e8 h1h2 d7c5 f4f5 b6b4 d2e3 c5d7 e3g3 b4e7 f5f6 e7f8 d1d4 g7g6 d4a4 a8a4 c3a4 d5d4 c2c3

I'll let the search continue, and report again later.

info depth 60 seldepth 89 multipv 1 score cp 68 nodes 576843828167 nps 30663264 hashfull 1000 tbhits 0 time 18812212 pv c1f4 f8e7 d1d2 e8g8 e1c1 c6d4 d2d4 a7a6 e4e5 d6e5 d4e5 e7d6 e5d4 d6f4 d4f4 d8e7 f1c4 c8e6 h1e1 a8d8 c4e6 f7e6 f4e3 d8d1 c3d1 f8e8 d1c3 e7d6 g2g3 d6c6 f2f3 g8f7 h2h4 f6d5 c3d5 e6d5 e3f4 f7g8 e1e8 c6e8 b2b3 e8d7 a2a4 a6a5 g3g4 h7h6 c1b1 c7c6 f4e5 g8h7 b1c1 d7f7 f3f4 f7f6 e5f6 g7f6 h4h5 h7g8 c1d2 g8g7 d2d3 c6c5 d3e3 g7g8 e3e2 g8f7 e2f3 b7b6 f3g3 f7g8

Thanks Louis, so Bf4 is Stockfish's choice at depth 60. Was Bb5 considered as the best move at lower depths?

[zullil]

I have a question to a similar position:

[d]r1bqkb1r/ppp2ppp/2np1n2/8/3NP3/2N5/PPP2PPP/R1BQKB1R w KQkq - 0 6

This can happen out of Philidor or Scotch, so that's very common position on an amateur level. So what's White's best way to continue? I guess many good moves here. Leela favours Bb5, while Stockfish switches between Bf4, Be2, Nxc6. Can someone with a strong CPU run Stockfish for a long time in this position?

As you say, Stockfish's choices vary considerably from iteration to iteration. Here's what Stockfish-dev offers at depth 50:

info depth 50 seldepth 83 multipv 1 score cp 84 nodes 95191474646 nps 29182257 hashfull 1000 tbhits 0 time 3261964 pv d4c6 b7c6 f1d3 f8e7 e1g1 e8g8 f2f4 a7a5 h2h3 c8b7 c1d2 f6d7 d1f3 e7f6 a1d1 f6d4 d2e3 d4e3 f3e3 d7c5 d3c4 a5a4 f1e1 d8b8 e4e5 b7c8 e3d4 c8e6 b2b3 a4b3 a2b3 e6c4 d4c4 b8b6 g1h1 d6d5 c4d4 c5d7 d4d2 f8e8 h1h2 d7c5 f4f5 b6b4 d2e3 c5d7 e3g3 b4e7 f5f6 e7f8 d1d4 g7g6 d4a4 a8a4 c3a4 d5d4 c2c3

I'll let the search continue, and report again later.

info depth 60 seldepth 89 multipv 1 score cp 68 nodes 576843828167 nps 30663264 hashfull 1000 tbhits 0 time 18812212 pv c1f4 f8e7 d1d2 e8g8 e1c1 c6d4 d2d4 a7a6 e4e5 d6e5 d4e5 e7d6 e5d4 d6f4 d4f4 d8e7 f1c4 c8e6 h1e1 a8d8 c4e6 f7e6 f4e3 d8d1 c3d1 f8e8 d1c3 e7d6 g2g3 d6c6 f2f3 g8f7 h2h4 f6d5 c3d5 e6d5 e3f4 f7g8 e1e8 c6e8 b2b3 e8d7 a2a4 a6a5 g3g4 h7h6 c1b1 c7c6 f4e5 g8h7 b1c1 d7f7 f3f4 f7f6 e5f6 g7f6 h4h5 h7g8 c1d2 g8g7 d2d3 c6c5 d3e3 g7g8 e3e2 g8f7 e2f3 b7b6 f3g3 f7g8

Thanks Louis, so Bf4 is Stockfish's choice at depth 60. Was Bb5 considered as the best move at lower depths?

You're welcome. A review of the search from depth 30 onward showed the following sequence of first choices: f1b5, f1e2, c1f4, d4c6, and back to c1f4.

Here's depth 65, and I've stopped the search there:

info depth 65 seldepth 113 multipv 1 score cp 94 nodes 1767923062777 nps 30891718 hashfull 1000 tbhits 0 time 57229677 pv c1f4 f8e7 d1d2 e8g8 e1c1 c6d4 d2d4 a7a6 e4e5 d6e5 d4e5 e7d6 e5d4 d6f4 d4f4 d8e7 f1c4 c8e6 h1e1 b7b5 c4e6 f7e6 a2a3 a8d8 f4e3 g8f7 f2f4 d8d6 c1b1 f8d8 f4f5 d8e8 d1d6 e7d6 c3e4 f6e4 e3e4 e6e5 e4f3 f7f8 h2h3 h7h6 g2g4 d6f6 e1e4 f6d6 b1c1 f8f7 e4e1 f7f8 c2c3 c7c5 f3e4 c5c4 e1d1 d6e7 c1b1 e8d8 d1e1 e7d6 b1a2 d6d5 e4e5 d5e5 e1e5 d8d3 h3h4 d3h3 h4h5 h3h4 a3a4 h4g4 a4b5 a6b5 e5b5 g4g5

[zullil]

I have a question to a similar position:

[d]r1bqkb1r/ppp2ppp/2np1n2/8/3NP3/2N5/PPP2PPP/R1BQKB1R w KQkq - 0 6

This can happen out of Philidor or Scotch, so that's very common position on an amateur level. So what's White's best way to continue? I guess many good moves here. Leela favours Bb5, while Stockfish switches between Bf4, Be2, Nxc6. Can someone with a strong CPU run Stockfish for a long time in this position?

Yes, Lc0 (network ) preferred Bb5 ... until it crashed. There's a bug in the code somewhere ...

info depth 25 seldepth 67 time 1898107 nodes 34661891 score cp 45 wdl 491 315 194 nps 18285 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 e8g8 b2b3 a7a5 d1f3 d8d7 c1b2 d7g4 f3d3 g4g6 a1e1 d6d5 e1e2 d5e4 c3e4 f6e4 e2e4 e7c5 g1h1 a8e8 f1e1 e8e4 e1e4 h7h5 h2h3 h5h4 d3e2 g6g3 b2g7 g8g7 e4e5 g3f4 e5c5 f8e8 e2e8 f4c1 h1h2 c1f4 h2g1 f4d4 g1f1 d4c5 e8e4 f7f5 e4h4 c5c2 h4e7 g7g6 e7e6 g6g7 e6e5 g7g6 e5a5 c2d3 f1f2 d3d4
Segmentation fault

Code: Select all

$ cat /var/log/syslog | grep lc0
May  3 12:03:59 LZsT7910 kernel: [ 2328.013612] lc0[3228]: segfault at 4000044 ip 00005613071ae73e sp 00007fc381ffdf90 error 6 in lc0[56130714b000+209000]

error 6 --- No page found for user-mode write access.

[Karol Majewski]

Yes, Lc0 (network ) preferred Bb5 ... until it crashed. There's a bug in the code somewhere ...

info depth 25 seldepth 67 time 1898107 nodes 34661891 score cp 45 wdl 491 315 194 nps 18285 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 e8g8 b2b3 a7a5 d1f3 d8d7 c1b2 d7g4 f3d3 g4g6 a1e1 d6d5 e1e2 d5e4 c3e4 f6e4 e2e4 e7c5 g1h1 a8e8 f1e1 e8e4 e1e4 h7h5 h2h3 h5h4 d3e2 g6g3 b2g7 g8g7 e4e5 g3f4 e5c5 f8e8 e2e8 f4c1 h1h2 c1f4 h2g1 f4d4 g1f1 d4c5 e8e4 f7f5 e4h4 c5c2 h4e7 g7g6 e7e6 g6g7 e6e5 g7g6 e5a5 c2d3 f1f2 d3d4
Segmentation fault

Code: Select all

$ cat /var/log/syslog | grep lc0
May  3 12:03:59 LZsT7910 kernel: [ 2328.013612] lc0[3228]: segfault at 4000044 ip 00005613071ae73e sp 00007fc381ffdf90 error 6 in lc0[56130714b000+209000]

error 6 --- No page found for user-mode write access.

Well, it's not possible to do infinite analysis with Leela. You are always limited by RAM. So once the size of the MCTS tree is bigger than the amount of available memory on you CPU - Leela crashes. To avoid this, in lc0.config set --ram-limit-mb=xxxxx where xxxxx is amount of memory for analysis. This will stop analysis once xxxxx amount of memory is taken - without crashing.

In our case, 34661891 nodes has been visited. Which is quite a lot.

[zullil]

Yes, Lc0 (network ) preferred Bb5 ... until it crashed. There's a bug in the code somewhere ...

info depth 25 seldepth 67 time 1898107 nodes 34661891 score cp 45 wdl 491 315 194 nps 18285 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 e8g8 b2b3 a7a5 d1f3 d8d7 c1b2 d7g4 f3d3 g4g6 a1e1 d6d5 e1e2 d5e4 c3e4 f6e4 e2e4 e7c5 g1h1 a8e8 f1e1 e8e4 e1e4 h7h5 h2h3 h5h4 d3e2 g6g3 b2g7 g8g7 e4e5 g3f4 e5c5 f8e8 e2e8 f4c1 h1h2 c1f4 h2g1 f4d4 g1f1 d4c5 e8e4 f7f5 e4h4 c5c2 h4e7 g7g6 e7e6 g6g7 e6e5 g7g6 e5a5 c2d3 f1f2 d3d4
Segmentation fault

Code: Select all

$ cat /var/log/syslog | grep lc0
May  3 12:03:59 LZsT7910 kernel: [ 2328.013612] lc0[3228]: segfault at 4000044 ip 00005613071ae73e sp 00007fc381ffdf90 error 6 in lc0[56130714b000+209000]

error 6 --- No page found for user-mode write access.

Well, it's not possible to do infinite analysis with Leela. You are always limited by RAM. So once the size of the MCTS tree is bigger than the amount of available memory on you CPU - Leela crashes. To avoid this, in lc0.config set --ram-limit-mb=xxxxx where xxxxx is amount of memory for analysis. This will stop analysis once xxxxx amount of memory is taken - without crashing.

In our case, 34661891 nodes has been visited. Which is quite a lot.

Yes, thanks. But I'm aware of all this. I have 128 GB of RAM, and set a limit of 102400 MB (=100 GB).

For this search, I set NNCacheSize to 10000000. According to Lc0's own log file for the search:

Code: Select all

RAM limit 102400MB. Cache takes 3200MB. Remaining memory is enough for 496000000 nodes.

So, when the crash occurred, plenty of RAM should certainly have been available.

[Henk]

Hi hi hi. One should store the tree in a file or database. That will perform 1000 times slower or so I don't know.

Looks like storing the tree limits these monte carlo tree expansion methods. That's another reason why alpha beta still popular.
Although hash table cost much memory too.

Maybe prune or clear up the monte carlo tree when lacking memory. (Or build a summary??)

I think I can run Skipper's analysis on my computer for days. But never tried. O wait maybe memory leaks in Skippers source code. I don't know.

[zullil]

Yes, Lc0 (network ) preferred Bb5 ... until it crashed. There's a bug in the code somewhere ...

info depth 25 seldepth 67 time 1898107 nodes 34661891 score cp 45 wdl 491 315 194 nps 18285 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 e8g8 b2b3 a7a5 d1f3 d8d7 c1b2 d7g4 f3d3 g4g6 a1e1 d6d5 e1e2 d5e4 c3e4 f6e4 e2e4 e7c5 g1h1 a8e8 f1e1 e8e4 e1e4 h7h5 h2h3 h5h4 d3e2 g6g3 b2g7 g8g7 e4e5 g3f4 e5c5 f8e8 e2e8 f4c1 h1h2 c1f4 h2g1 f4d4 g1f1 d4c5 e8e4 f7f5 e4h4 c5c2 h4e7 g7g6 e7e6 g6g7 e6e5 g7g6 e5a5 c2d3 f1f2 d3d4
Segmentation fault

Code: Select all

$ cat /var/log/syslog | grep lc0
May  3 12:03:59 LZsT7910 kernel: [ 2328.013612] lc0[3228]: segfault at 4000044 ip 00005613071ae73e sp 00007fc381ffdf90 error 6 in lc0[56130714b000+209000]

error 6 --- No page found for user-mode write access.

Well, it's not possible to do infinite analysis with Leela. You are always limited by RAM. So once the size of the MCTS tree is bigger than the amount of available memory on you CPU - Leela crashes. To avoid this, in lc0.config set --ram-limit-mb=xxxxx where xxxxx is amount of memory for analysis. This will stop analysis once xxxxx amount of memory is taken - without crashing.

In our case, 34661891 nodes has been visited. Which is quite a lot.

Tried again, and this time no crash.

Lc0-dev (network 384x30-swa-3350000), with 100M nodes indeed prefers Bb5:

Code: Select all

0507 08:17:40.399343 140089188962752 ../../src/chess/uciloop.cc:135] >> ucinewgame
0507 08:17:40.400541 140089188962752 ../../src/neural/loader.cc:230] Found pb network file: ./384x30-swa-3350000.pb
0507 08:17:40.796115 140089188962752 ../../src/neural/factory.cc:91] Creating backend [cudnn-auto]...
0507 08:17:40.836600 140089188962752 ../../src/neural/cuda/network_cudnn.cc:1039] Switching to [cudnn-fp16]...
0507 08:17:41.004662 140089188962752 ../../src/neural/cuda/network_cudnn.cc:913] CUDA Runtime version: 10.2.0
0507 08:17:41.004763 140089188962752 ../../src/neural/cuda/network_cudnn.cc:926] Cudnn version: 7.6.5
0507 08:17:41.004777 140089188962752 ../../src/neural/cuda/network_cudnn.cc:936] Latest version of CUDA supported by the driver: 10.2.0
0507 08:17:41.004971 140089188962752 ../../src/neural/cuda/network_cudnn.cc:944] GPU: GeForce RTX 2080 Ti
0507 08:17:41.004989 140089188962752 ../../src/neural/cuda/network_cudnn.cc:945] GPU memory: 10.7534 Gb
0507 08:17:41.005029 140089188962752 ../../src/neural/cuda/network_cudnn.cc:947] GPU clock frequency: 1635 MHz
0507 08:17:41.005043 140089188962752 ../../src/neural/cuda/network_cudnn.cc:948] GPU compute capability: 7.5
0507 08:18:04.104380 140089188962752 ../../src/chess/uciloop.cc:135] >> position fen r1bqkb1r/ppp2ppp/2np1n2/8/3NP3/2N5/PPP2PPP/R1BQKB1R w KQkq - 0 6
0507 08:18:20.464094 140089188962752 ../../src/chess/uciloop.cc:135] >> go nodes 100000000
0507 08:18:20.465201 140089188962752 ../../src/mcts/stoppers/stoppers.cc:104] RAM limit 102400MB. Cache takes 3200MB. Remaining memory is enough for 496000000 nodes.

info depth 29 seldepth 90 time 6271819 nodes 100000058 score cp 43 wdl 486 314 200 nps 15986 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 f6d7 b2b3 e8g8 c1b2 e7f6 d1d3 c6c5 g1h1 f6d4 a1e1 f8e8 b2c1 d4c3 d3c3 e8e6 e1e2 h7h6 h2h3 d8h4 c3f3 a8e8 c1b2 f7f5 e4e5 d7f6 h1h2 f6g4 h2g1 g4f6 f3d3 c7c6 g1h2 f6g4 h2g1 g4f6 d3c4 h4h5 e2e3 f6e4 e5d6 e4d6 c4c3 h5f7 e3e6 e8e6 c3c5 d6e4 c5e3 e6d6 f1e1 h6h5 b2c3 h5h4 e1c1 c6c5 c3e1 f7f6 g1h2 f6b2 a2a4 g8h7 a4a5 a7a6

Code: Select all

0507 10:02:35.923984 140088423069440 ../../src/mcts/stoppers/stoppers.cc:64] Stopped search: Reached visits limit: 100000058>=100000000
0507 10:02:35.924367 140088423069440 ../../src/chess/uciloop.cc:222] << info depth 29 seldepth 90 time 6271819 nodes 100000058 score cp 43 wdl 486 314 200 nps 15986 tbhits 0 pv f1b5 c8d7 e1g1 f8e7 d4c6 d7c6 b5c6 b7c6 f2f4 f6d7 b2b3 e8g8 c1b2 e7f6 d1d3 c6c5 g1h1 f6d4 a1e1 f8e8 b2c1 d4c3 d3c3 e8e6 e1e2 h7h6 h2h3 d8h4 c3f3 a8e8 c1b2 f7f5 e4e5 d7f6 h1h2 f6g4 h2g1 g4f6 f3d3 c7c6 g1h2 f6g4 h2g1 g4f6 d3c4 h4h5 e2e3 f6e4 e5d6 e4d6 c4c3 h5f7 e3e6 e8e6 c3c5 d6e4 c5e3 e6d6 f1e1 h6h5 b2c3 h5h4 e1c1 c6c5 c3e1 f7f6 g1h2 f6b2 a2a4 g8h7 a4a5 a7a6

0507 10:02:35.924664 140088423069440 ../../src/mcts/search.cc:374] === Move stats:
0507 10:02:35.924679 140088423069440 ../../src/mcts/search.cc:375] d4e6  (767 ) N:     728 (+ 0) (P:  0.49%) (WL: -0.89093) (D:  0.024) (M:  3.5) (Q: -0.89093) (U: 1.77895) (S:  0.88801) (V:  -.----) 
0507 10:02:35.924686 140088423069440 ../../src/mcts/search.cc:375] e1d2  (105 ) N:     730 (+ 0) (P:  0.50%) (WL: -0.89700) (D:  0.033) (M:  3.7) (Q: -0.89700) (U: 1.78714) (S:  0.89014) (V:  -.----) 
0507 10:02:35.924692 140088423069440 ../../src/mcts/search.cc:375] d1h5  (93  ) N:     738 (+ 0) (P:  0.50%) (WL: -0.90480) (D:  0.032) (M:  3.5) (Q: -0.90480) (U: 1.79909) (S:  0.89429) (V:  -.----) 
0507 10:02:35.924698 140088423069440 ../../src/mcts/search.cc:375] c1h6  (69  ) N:     799 (+ 0) (P:  0.53%) (WL: -0.87834) (D:  0.032) (M:  3.6) (Q: -0.87834) (U: 1.76055) (S:  0.88221) (V:  -.----) 
0507 10:02:35.924704 140088423069440 ../../src/mcts/search.cc:375] d1f3  (88  ) N:     880 (+ 0) (P:  0.59%) (WL: -0.88463) (D:  0.030) (M:  3.7) (Q: -0.88463) (U: 1.77042) (S:  0.88579) (V:  -.----) 
0507 10:02:35.924724 140088423069440 ../../src/mcts/search.cc:375] d1g4  (91  ) N:     891 (+ 0) (P:  0.62%) (WL: -0.92014) (D:  0.031) (M:  3.6) (Q: -0.92014) (U: 1.82128) (S:  0.90114) (V:  -.----) 
0507 10:02:35.924731 140088423069440 ../../src/mcts/search.cc:375] c1d2  (57  ) N:     893 (+ 0) (P:  0.60%) (WL: -0.88063) (D:  0.030) (M:  3.9) (Q: -0.88063) (U: 1.76379) (S:  0.88316) (V:  -.----) 
0507 10:02:35.924736 140088423069440 ../../src/mcts/search.cc:375] f1d3  (134 ) N:     949 (+ 0) (P:  0.64%) (WL: -0.89012) (D:  0.031) (M:  3.9) (Q: -0.89012) (U: 1.77834) (S:  0.88823) (V:  -.----) 
0507 10:02:35.924741 140088423069440 ../../src/mcts/search.cc:375] f1a6  (143 ) N:     966 (+ 0) (P:  0.65%) (WL: -0.88438) (D:  0.031) (M:  4.0) (Q: -0.88438) (U: 1.76891) (S:  0.88453) (V:  -.----) 
0507 10:02:35.924747 140088423069440 ../../src/mcts/search.cc:375] d1e2  (82  ) N:    1008 (+ 0) (P:  0.68%) (WL: -0.88505) (D:  0.029) (M:  4.0) (Q: -0.88505) (U: 1.77100) (S:  0.88595) (V:  -.----) 
0507 10:02:35.924752 140088423069440 ../../src/mcts/search.cc:375] e4e5  (796 ) N:    1229 (+ 0) (P:  0.56%) (WL: -0.49557) (D:  0.184) (M:  6.7) (Q: -0.49557) (U: 1.19044) (S:  0.69487) (V:  -.----) 
0507 10:02:35.924758 140088423069440 ../../src/mcts/search.cc:375] d4f5  (763 ) N:    1583 (+ 0) (P:  0.72%) (WL: -0.49475) (D:  0.185) (M:  8.4) (Q: -0.49475) (U: 1.19000) (S:  0.69525) (V:  -.----) 
0507 10:02:35.924763 140088423069440 ../../src/mcts/search.cc:375] e1e2  (106 ) N:    1719 (+ 0) (P:  0.59%) (WL: -0.30538) (D:  0.261) (M:  6.9) (Q: -0.30538) (U: 0.90799) (S:  0.60261) (V:  -.----) 
0507 10:02:35.924768 140088423069440 ../../src/mcts/search.cc:375] g2g4  (378 ) N:    1777 (+ 0) (P:  0.60%) (WL: -0.29183) (D:  0.226) (M:  7.6) (Q: -0.29183) (U: 0.88799) (S:  0.59616) (V:  -.----) 
0507 10:02:35.924774 140088423069440 ../../src/mcts/search.cc:375] c3b1  (467 ) N:    1859 (+ 0) (P:  0.68%) (WL: -0.34177) (D:  0.224) (M: 10.0) (Q: -0.34177) (U: 0.96180) (S:  0.62003) (V:  -.----) 
0507 10:02:35.924779 140088423069440 ../../src/mcts/search.cc:375] c3e2  (475 ) N:    1889 (+ 0) (P:  0.59%) (WL: -0.25245) (D:  0.282) (M:  7.4) (Q: -0.25245) (U: 0.82925) (S:  0.57680) (V:  -.----) 
0507 10:02:35.924784 140088423069440 ../../src/mcts/search.cc:375] c3b5  (489 ) N:    2151 (+ 0) (P:  0.65%) (WL: -0.22738) (D:  0.248) (M:  8.1) (Q: -0.22738) (U: 0.79182) (S:  0.56444) (V:  -.----) 
0507 10:02:35.924790 140088423069440 ../../src/mcts/search.cc:375] b2b4  (234 ) N:    2690 (+ 0) (P:  0.66%) (WL: -0.13110) (D:  0.342) (M:  8.8) (Q: -0.13110) (U: 0.64891) (S:  0.51781) (V:  -.----) 
0507 10:02:35.924795 140088423069440 ../../src/mcts/search.cc:375] c3a4  (483 ) N:    2919 (+ 0) (P:  0.72%) (WL: -0.13408) (D:  0.297) (M:  9.5) (Q: -0.13408) (U: 0.65328) (S:  0.51920) (V:  -.----) 
0507 10:02:35.924801 140088423069440 ../../src/mcts/search.cc:375] h1g1  (177 ) N:    4448 (+ 0) (P:  0.83%) (WL: -0.02343) (D:  0.316) (M:  9.6) (Q: -0.02343) (U: 0.48876) (S:  0.46533) (V:  -.----) 
0507 10:02:35.924806 140088423069440 ../../src/mcts/search.cc:375] d4b5  (759 ) N:    6384 (+ 0) (P:  0.94%) (WL:  0.04555) (D:  0.345) (M: 11.2) (Q:  0.04555) (U: 0.38606) (S:  0.43162) (V:  -.----) 
0507 10:02:35.924812 140088423069440 ../../src/mcts/search.cc:375] c3d5  (491 ) N:    7826 (+ 0) (P:  1.04%) (WL:  0.06873) (D:  0.348) (M:  9.6) (Q:  0.06873) (U: 0.35154) (S:  0.42027) (V:  -.----) 
0507 10:02:35.924817 140088423069440 ../../src/mcts/search.cc:375] h2h4  (403 ) N:   12938 (+ 0) (P:  1.21%) (WL:  0.13964) (D:  0.279) (M: 10.0) (Q:  0.13964) (U: 0.24606) (S:  0.38571) (V:  -.----) 
0507 10:02:35.924822 140088423069440 ../../src/mcts/search.cc:375] d4e2  (745 ) N:   13272 (+ 0) (P:  0.95%) (WL:  0.17821) (D:  0.267) (M:  7.2) (Q:  0.17821) (U: 0.18867) (S:  0.36688) (V:  -.----) 
0507 10:02:35.924828 140088423069440 ../../src/mcts/search.cc:375] d4b3  (747 ) N:   13297 (+ 0) (P:  1.33%) (WL:  0.12779) (D:  0.308) (M:  7.1) (Q:  0.12779) (U: 0.26369) (S:  0.39148) (V:  -.----) 
0507 10:02:35.924837 140088423069440 ../../src/mcts/search.cc:375] d4f3  (751 ) N:   13302 (+ 0) (P:  1.28%) (WL:  0.13395) (D:  0.328) (M:  6.6) (Q:  0.13395) (U: 0.25452) (S:  0.38847) (V:  -.----) 
0507 10:02:35.924842 140088423069440 ../../src/mcts/search.cc:375] a1b1  (0   ) N:   13339 (+ 0) (P:  1.08%) (WL:  0.16101) (D:  0.338) (M:  7.4) (Q:  0.16101) (U: 0.21425) (S:  0.37526) (V:  -.----) 
0507 10:02:35.924848 140088423069440 ../../src/mcts/search.cc:375] b2b3  (230 ) N:   13370 (+ 0) (P:  1.46%) (WL:  0.11165) (D:  0.291) (M: 10.4) (Q:  0.11165) (U: 0.28766) (S:  0.39931) (V:  -.----) 
0507 10:02:35.924853 140088423069440 ../../src/mcts/search.cc:375] d1d2  (81  ) N:   14470 (+ 0) (P:  1.28%) (WL:  0.14820) (D:  0.302) (M:  9.2) (Q:  0.14820) (U: 0.23328) (S:  0.38149) (V:  -.----) 
0507 10:02:35.924859 140088423069440 ../../src/mcts/search.cc:375] f2f4  (351 ) N:   17061 (+ 0) (P:  1.04%) (WL:  0.19715) (D:  0.282) (M:  9.9) (Q:  0.19715) (U: 0.16044) (S:  0.35760) (V:  -.----) 
0507 10:02:35.924864 140088423069440 ../../src/mcts/search.cc:375] a2a3  (204 ) N:   22852 (+ 0) (P:  1.60%) (WL:  0.18106) (D:  0.314) (M:  8.4) (Q:  0.18106) (U: 0.18441) (S:  0.36547) (V:  -.----) 
0507 10:02:35.924869 140088423069440 ../../src/mcts/search.cc:375] c1g5  (67  ) N:   27052 (+ 0) (P:  2.30%) (WL:  0.15415) (D:  0.276) (M:  9.3) (Q:  0.15415) (U: 0.22445) (S:  0.37860) (V:  -.----) 
0507 10:02:35.924875 140088423069440 ../../src/mcts/search.cc:375] a2a4  (207 ) N:   27398 (+ 0) (P:  1.44%) (WL:  0.21212) (D:  0.319) (M:  8.4) (Q:  0.21212) (U: 0.13818) (S:  0.35030) (V:  -.----) 
0507 10:02:35.924880 140088423069440 ../../src/mcts/search.cc:375] f2f3  (346 ) N:   35531 (+ 0) (P:  2.46%) (WL:  0.18234) (D:  0.294) (M: 10.3) (Q:  0.18234) (U: 0.18249) (S:  0.36483) (V:  -.----) 
0507 10:02:35.924885 140088423069440 ../../src/mcts/search.cc:375] c1e3  (63  ) N:   40527 (+ 0) (P:  1.97%) (WL:  0.21893) (D:  0.296) (M:  9.4) (Q:  0.21893) (U: 0.12805) (S:  0.34697) (V:  -.----) 
0507 10:02:35.924891 140088423069440 ../../src/mcts/search.cc:375] g2g3  (374 ) N:   42310 (+ 0) (P:  2.34%) (WL:  0.20719) (D:  0.297) (M:  9.1) (Q:  0.20719) (U: 0.14551) (S:  0.35269) (V:  -.----) 
0507 10:02:35.924896 140088423069440 ../../src/mcts/search.cc:375] h2h3  (400 ) N:   69813 (+ 0) (P:  4.01%) (WL:  0.20318) (D:  0.300) (M: 17.3) (Q:  0.20318) (U: 0.15143) (S:  0.35461) (V:  -.----) 
0507 10:02:35.924901 140088423069440 ../../src/mcts/search.cc:375] d1d3  (86  ) N:   69944 (+ 0) (P:  2.86%) (WL:  0.23258) (D:  0.249) (M: 10.3) (Q:  0.23258) (U: 0.10768) (S:  0.34026) (V:  -.----) 
0507 10:02:35.924907 140088423069440 ../../src/mcts/search.cc:375] f1c4  (139 ) N:   78515 (+ 0) (P:  3.11%) (WL:  0.23478) (D:  0.317) (M:  8.9) (Q:  0.23478) (U: 0.10440) (S:  0.33919) (V:  -.----) 
0507 10:02:35.924912 140088423069440 ../../src/mcts/search.cc:375] f1e2  (130 ) N:  198962 (+ 0) (P:  7.51%) (WL:  0.23800) (D:  0.281) (M: 12.4) (Q:  0.23800) (U: 0.09948) (S:  0.33748) (V:  -.----) 
0507 10:02:35.924918 140088423069440 ../../src/mcts/search.cc:375] d4c6  (765 ) N:  263944 (+ 0) (P: 11.77%) (WL:  0.22579) (D:  0.297) (M: 12.2) (Q:  0.22579) (U: 0.11755) (S:  0.34334) (V:  -.----) 
0507 10:02:35.924923 140088423069440 ../../src/mcts/search.cc:375] c1f4  (65  ) N: 3237123 (+ 0) (P:  9.01%) (WL:  0.27808) (D:  0.259) (M: 16.9) (Q:  0.27808) (U: 0.00733) (S:  0.28541) (V:  -.----) 
0507 10:02:35.924928 140088423069440 ../../src/mcts/search.cc:375] f1b5  (141 ) N: 95729981 (+42) (P: 25.61%) (WL:  0.28571) (D:  0.314) (M: 22.2) (Q:  0.28571) (U: 0.00071) (S:  0.28642) (V:  -.----) 

[Karol Majewski]

Thanks. Note that Bf4, which is Stockfish's choice, is a clear second here. I think it's safe to say that the best move is one of the two. I wonder why Stockfish doesn't like Bb5 that much. To find out, we should put these into Stockfish:

position startpos moves e2e4 e7e5 g1f3 b8c6 b1c3 g8f6 d2d4 e5d4 f3d4 d7d6 c1f4 f8e7
go depth x

vs

position startpos moves e2e4 e7e5 g1f3 b8c6 b1c3 g8f6 d2d4 e5d4 f3d4 d7d6 f1b5 c8d7
go depth x

And compare score cp after depth x. Of course x should be 60 or higher.