ZH : more wins by opperwezen (may 2017)

[Vinvin20]

https://fr.lichess.org/vYwqfgZU https://fr.lichess.org/izq1HWJ6 https://fr.lichess.org/1duaQvZU https://fr.lichess.org/BtJGUDyW https://fr.lichess.org/eCH3TZEf And even one 1+0 : https://fr.lichess.org/0P4BY7oA

[ddugovic]

In the first game SF blundered spectacularly, and this is still reproducible:

setoption name UCI_Variant value crazyhouse
position fen r4r1k/p1pn1ppp/1p6/3QP3/5b1P/7P/PPP1PPB1/R4RK1/PBPQnnbn b - - 33 17
go movetime 1000

I don't understand how Stockfish gains so many Elo in self-play, yet in blitz misses mating nets like 18. P@f6 and the game continuation.

[ddugovic]

Ah, apparently B@h2+! is Black's best move in that first game (tested with 87fbdfd):

setoption name UCI_Variant value crazyhouse
setoption name Threads value 2
setoption name MultiPV value 2
position fen r4r1k/p1pn1ppp/1p6/3QP3/5b1P/7P/PPP1PPB1/R4RK1/PBPQnnbn b - - 33 17
go movetime 4000

info depth 13 seldepth 21 multipv 1 score cp -14 nodes 1614502 nps 1016049 hashfull 678 tbhits 0 time 1589 pv B@e6 d5e4 f4e5 P@e7 f8g8 Q@h5 N@g6 e4a8 g8a8 R@d8 a8d8 e7d8q R@f8 d8f8 d7f8 P@g3
info depth 13 seldepth 21 multipv 2 score cp -54 nodes 1614502 nps 1016049 hashfull 678 tbhits 0 time 1589 pv d7e5 P@g3 B@c6 d5c6 e5c6 g3f4 Q@g6 P@g3 a8e8 B@c3 N@d4 B@f3 e8e2 f3c6 e2c2 c3d4 N@e2 g1h1 e2d4
info depth 14 seldepth 27 multipv 1 score cp 132 nodes 4276032 nps 1063689 hashfull 993 tbhits 0 time 4020 pv B@h2
info depth 13 seldepth 27 multipv 2 score cp -54 nodes 4276032 nps 1063689 hashfull 993 tbhits 0 time 4020 pv d7e5 P@g3
bestmove B@h2 ponder g1h1

[Vinvin20]

I don't understand how Stockfish gains so many Elo in self-play, yet in blitz misses mating nets like 18. P@f6 and the game continuation.

You can make good results in "self-play" if the opponent is not able to exploit this weakness. That's the weak spot in "self-play". (it's not exactly "self-play" because the opponent is a bit different)

[Vinvin20]

Some more recent wins by humans : https://lichess.org/B3QVomKy <- 10.Bxf5? -> Nxf2 seems better https://lichess.org/TUEPRIng https://lichess.org/Gu06q6Dn

[ddugovic]

Thanks @Vinvin20 . I think r4r1k/p1pn1ppp/1p6/3QP3/5b1P/7P/PPP1PPB1/R4RK1/PBPQnnbn b - - 33 17 is an excellent test position, as not only did SF blunder during the game but even during analysis it struggles to find a better move and basically walks into a mating net.

[ddugovic]

After some more experimentation we find a good move earlier... I look forward to trying this plus the result of the tuning session:

setoption name UCI_Variant value crazyhouse
setoption name Threads value 2
setoption name MultiPV value 2
position fen r4r1k/p1pn1ppp/1p6/3QP3/5b1P/7P/PPP1PPB1/R4RK1/PBPQnnbn b - - 33 17
go movetime 3000

info depth 9 seldepth 15 multipv 1 score cp 163 nodes 167573 nps 957560 tbhits 0 time 175 pv d7e5 B@g3 B@e6 d5a8 f8a8 g3f4 Q@g6 f4e5 e6h3 P@g5
info depth 9 seldepth 15 multipv 2 score cp 16 nodes 167573 nps 957560 tbhits 0 time 175 pv B@h2 g1h1 d7e5 P@e7 N@f5 e7f8q a8f8 R@a8 N@g3 f2g3 f5g3 h1h2 g3f1 h2g1 f4e3 g1f1
info depth 10 seldepth 20 multipv 1 score cp 68 nodes 298821 nps 1063419 tbhits 0 time 281 pv d7e5 P@g3 B@c6 d5c6 e5c6 g2c6 N@d4 Q@g4 Q@g6 g4g6 d4e2 g1h1 h7g6
info depth 10 seldepth 20 multipv 2 score cp -98 nodes 298821 nps 1063419 tbhits 0 time 281 pv B@h2 g1h1 d7e5 P@g3 P@f3 g3f4 f3g2 d5g2 B@c6 P@d5 c6d5 g2d5 h2f4 P@g3
info depth 11 seldepth 20 multipv 1 score cp 28 nodes 455387 nps 1071498 tbhits 0 time 425 pv d7e5 P@g3 B@c6 d5c6 e5c6 g3f4 N@d4 B@d3 f7f5 P@b7 Q@g6 b7a8q f8a8 g1h1 P@e4
info depth 11 seldepth 20 multipv 2 score cp -90 nodes 455387 nps 1071498 tbhits 0 time 425 pv B@h2 g1h1 d7e5 P@g3 P@f3 g3f4 h2f4 P@f6 g7f6 g2f3 N@f5 Q@g2 P@g3 f3h5 c7c6
info depth 12 seldepth 24 multipv 1 score cp 85 nodes 1292243 nps 1120765 hashfull 588 tbhits 0 time 1153 pv B@h2 g1h1 N@f5 P@g3 h2g3 P@h6 g3f2 h6g7 f5g7 P@h2 P@f3 f1f2 f3g2 d5g2 f4e5 P@e7
info depth 12 seldepth 24 multipv 2 score cp -32 upperbound nodes 1292243 nps 1120765 hashfull 588 tbhits 0 time 1153 pv d7e5 P@g3 B@c6 d5c6 e5c6 g3f4 Q@g6 B@d3 f7f5 P@g3 P@e4 d3c4 c6d4 Q@e5 d4c2
info depth 13 seldepth 24 multipv 1 score cp 40 nodes 3427375 nps 1137151 hashfull 963 tbhits 0 time 3014 pv B@h2 g1h1 N@f5 P@g3 h2g3 P@h6 g3f2 h6g7 f5g7 P@h2 N@e3 d5d7 f4h2 f1f2 h2e5 B@f6 N@g3 h1h2 g3f1 h2g1 P@h2 g1h1
info depth 13 seldepth 24 multipv 2 score cp -188 upperbound nodes 3427375 nps 1137151 hashfull 963 tbhits 0 time 3014 pv d7e5 P@h6
bestmove B@h2 ponder g1h1