Stockfish: Crazyhouse search performance

Created on 9 Jan 2019  路  9Comments  路  Source: ddugovic/Stockfish

Making engines calculate like humans might not be solvable, but perhaps there's a problem or room for improvement with search as suggested by @QueensGambit at https://github.com/ddugovic/Stockfish/issues/158#issuecomment-452755029:

setoption name UCI_Variant value crazyhouse
position fen rnbqk2r/ppp2ppp/4p3/3N4/3P4/5N2/PpPQ1PPP/R3KB1R/BNPpb w KQkq - 0 9
go infinite
info depth 1 seldepth 2 multipv 1 score cp 534 nodes 409 nps 204500 tbhits 0 time 2 pv f1b5 P@c6
info depth 2 seldepth 3 multipv 1 score cp 702 nodes 691 nps 345500 tbhits 0 time 2 pv a1b1 d8d5 B@b5 d5b5 f1b5 P@c6 b1b2
info depth 3 seldepth 4 multipv 1 score cp 353 nodes 1010 nps 336666 tbhits 0 time 3 pv a1b1 d8d5 b1b2
info depth 4 seldepth 5 multipv 1 score cp 376 nodes 1346 nps 448666 tbhits 0 time 3 pv d5c7 d8c7 a1b1 B@c6 b1b2
info depth 5 seldepth 5 multipv 1 score cp 376 nodes 1513 nps 504333 tbhits 0 time 3 pv d5c7 d8c7 a1b1 B@c6 b1b2
info depth 6 seldepth 6 multipv 1 score cp 433 nodes 1702 nps 425500 tbhits 0 time 4 pv d5c7 d8c7 a1b1 B@c6
info depth 7 seldepth 9 multipv 1 score cp 224 nodes 4304 nps 614857 tbhits 0 time 7 pv a1b1 d8d5 b1b2 b8c6 B@a3 e6e5 d4e5
info depth 8 seldepth 9 multipv 1 score cp 171 nodes 6324 nps 632400 tbhits 0 time 10 pv a1b1 d8d5 b1b2 b8c6 B@a3 e6e5 f3e5 d5e4 f1e2 e4g2
info depth 9 seldepth 13 multipv 1 score cp 354 nodes 20636 nps 606941 tbhits 0 time 34 pv d5c7 d8c7 a1d1 N@e4 N@b5
info depth 10 seldepth 16 multipv 1 score cp 214 nodes 41325 nps 635769 tbhits 0 time 65 pv d5c7 d8c7 a1d1 B@c3 B@b4 c3d2 f3d2 b8d7 N@d6 c7d6 b4d6
info depth 11 seldepth 20 multipv 1 score cp 150 nodes 63264 nps 645551 tbhits 0 time 98 pv d5c7 d8c7 a1d1 B@f4 B@e5 f4d2 f3d2 c7c2 N@d6 e8e7 d6f5 c2f5 B@b4 e7d7 f1b5 P@c6
info depth 12 seldepth 19 multipv 1 score cp 311 nodes 119848 nps 627476 tbhits 0 time 191 pv d5c7 d8c7 a1d1 B@c3 B@e5 c3d2 f3d2 c7c2 N@d6 e8e7
info depth 13 seldepth 23 multipv 1 score cp 273 nodes 184979 nps 637858 tbhits 0 time 290 pv d5c7 d8c7 a1d1 B@f4 P@e3 N@a3 e3f4 a3c2 d2c2 c7c2 N@d6 e8e7 d6c8 h8c8 P@d6 e7d6 B@c5 c8c5 d4c5 d6c5
info depth 14 seldepth 25 multipv 1 score cp 368 nodes 372761 nps 636110 tbhits 0 time 586 pv d5c7 d8c7 a1d1 N@e4 N@b5 c7d7 d2e3 B@a5 P@d2 e8g8 e3e4 b8c6 P@h6 f7f5 h6g7 d7g7 e4e3
info depth 15 seldepth 32 multipv 1 score cp 133 nodes 1036324 nps 613209 hashfull 458 tbhits 0 time 1690 pv d5c7 d8c7 a1d1 B@c3 B@e5 c3d2 f3d2 c7c2 N@d6 e8e7 d6c8 h8c8 B@b4 P@d6 e5d6 e7d7 f1b5 P@c6 e1g1 c6b5 B@f3 N@c3 f3b7 N@e2 g1h1 c3d1 b7a8 b2b1q
info depth 16 seldepth 35 multipv 1 score cp 205 nodes 1097737 nps 615668 hashfull 478 tbhits 0 time 1783 pv d5c7 d8c7 a1d1 B@c3 B@e5 c3d2 f3d2 c7c2 N@d6 e8e7 B@b3 c2d2 e1d2 P@c3 d2e2 N@c1 e2f3 P@g4 f3g4 Q@g6 g4h3 c1b3
info depth 17 seldepth 31 multipv 1 score cp 87 upperbound nodes 1941637 nps 616784 hashfull 760 tbhits 0 time 3148 pv d5c7 d8c7
info depth 17 seldepth 32 multipv 1 score cp 39 upperbound nodes 2064980 nps 617333 hashfull 784 tbhits 0 time 3345 pv d5c7 d8c7
info depth 17 seldepth 32 multipv 1 score cp 77 lowerbound nodes 2542995 nps 612474 hashfull 859 tbhits 0 time 4152 pv a1b1
info depth 17 seldepth 32 multipv 1 score cp 158 lowerbound nodes 2638479 nps 611750 hashfull 869 tbhits 0 time 4313 pv a1b1
info depth 17 seldepth 32 multipv 1 score cp 117 nodes 2792786 nps 611246 hashfull 883 tbhits 0 time 4569 pv a1b1 e6d5 P@h6 N@e4 h6g7 h8g8 d2e2 B@a5 P@d2 b8c6 N@h5 P@e7 B@d3 c8g4 d3e4 g4h5 e4h7 g8g7 N@f6 e8f8 f6h5 g7h7 b1b2
info depth 18 seldepth 29 multipv 1 score cp 108 upperbound nodes 3081110 nps 610000 hashfull 917 tbhits 0 time 5051 pv a1b1 e6d5
info depth 18 seldepth 29 multipv 1 score cp 98 upperbound nodes 3370900 nps 609345 hashfull 936 tbhits 0 time 5532 pv a1b1 e6d5
info depth 18 seldepth 29 multipv 1 score cp 84 upperbound nodes 3548896 nps 609147 hashfull 944 tbhits 0 time 5826 pv a1b1 e6d5
info depth 18 seldepth 29 multipv 1 score cp 63 upperbound nodes 3649045 nps 610412 hashfull 949 tbhits 0 time 5978 pv a1b1 e6d5
info depth 18 seldepth 29 multipv 1 score cp 36 upperbound nodes 3787277 nps 611541 hashfull 953 tbhits 0 time 6193 pv a1b1 e6d5
info depth 18 seldepth 29 multipv 1 score cp 0 upperbound nodes 4101134 nps 611470 hashfull 966 tbhits 0 time 6707 pv a1b1 e6d5
info depth 18 seldepth 34 multipv 1 score cp 28 lowerbound nodes 4637037 nps 609334 hashfull 983 tbhits 0 time 7610 pv a1b1
info depth 18 seldepth 34 multipv 1 score cp 60 nodes 4993003 nps 608977 hashfull 986 tbhits 0 time 8199 pv a1b1 e6d5 P@h6 d8e7 N@e3 g7h6 f1b5 P@c6 e1g1 e8g8 b1b2 B@a3 P@f6 e7f6 B@e5 f6g6 e3d5 P@g7 e5g7 g8g7 P@f6 g7g8 d5e7 a3e7 f6e7 N@h3 g1h1 c6b5
info depth 19 seldepth 36 multipv 1 score cp 50 upperbound nodes 5886225 nps 606577 hashfull 996 tbhits 0 time 9704 pv a1b1 e6d5
info depth 19 seldepth 36 multipv 1 score cp 40 upperbound nodes 6271547 nps 605888 hashfull 998 tbhits 0 time 10351 pv a1b1 e6d5
info depth 19 seldepth 37 multipv 1 score cp 26 upperbound nodes 7373507 nps 604633 hashfull 1000 tbhits 0 time 12195 pv a1b1 e6d5
info depth 19 seldepth 37 multipv 1 score cp 6 upperbound nodes 7898092 nps 604800 hashfull 1000 tbhits 0 time 13059 pv a1b1 e6d5
info depth 19 seldepth 37 multipv 1 score cp -21 upperbound nodes 8451393 nps 605313 hashfull 1000 tbhits 0 time 13962 pv a1b1 e6d5
info depth 19 seldepth 39 multipv 1 score cp -57 upperbound nodes 9253042 nps 605011 hashfull 1000 tbhits 0 time 15294 pv a1b1 e6d5
info depth 19 seldepth 39 multipv 1 score cp -28 lowerbound nodes 9364810 nps 604805 hashfull 1000 tbhits 0 time 15484 pv a1b1
info depth 19 seldepth 41 multipv 1 score cp 34 lowerbound nodes 9613680 nps 604367 hashfull 1000 tbhits 0 time 15907 pv a1b1
info depth 19 seldepth 41 multipv 1 score cp 114 lowerbound nodes 9669649 nps 604541 hashfull 1000 tbhits 0 time 15995 pv a1b1
info depth 19 seldepth 41 multipv 1 score cp 73 nodes 10716484 nps 604699 hashfull 1000 tbhits 0 time 17722 pv a1d1 e6d5 P@h6 d8e7 B@e5 N@e4 h6g7 h8g8 P@f6 e4f6 N@e3 P@c6 c2c3 P@e4 d2b2 e4f3 e5f6 e7f6 N@h5 f6e7 g2f3

Most helpful comment

Hey, I agree with @ianfab that the transposition seems really unlikely to occur because if first Rd1 is played then taking the knight is almost forced by black. Consequently a later Nxc7 becomes impossible.

I'm not a SF expert, but for me it seems more like a potential minor issue in move ordering or soft pruning of some sort. From a purely objective point of view without looking at the position Nxc7 is easily the best move because it comes with check and white gets at least a pawn while opening a crucial back rank square for later attack. However, it also helps black developing a counter-attack with 10. B@c3 to follow which is a lot harder to evaluate correctly.

Nevertheless, the move Nxc7 doesn't seem like a bad move to me and it's not 100% certain that Rd1 is a better move in this position. The line 9. Nxc7 Qxc7 10. Rd1 B@c3 11. N@d6 Kf8 12. Nc4 is also promising for white.

All 9 comments

I wonder if there's a crazyhouse search bug? Rd1 followed by Nc7 could transpose into the same positions as Nc7 followed by Rd1.

$ cat test.txt - | ./stockfish | grep a1d1
info depth 19 seldepth 31 multipv 1 score cp 175 nodes 3588019 nps 605572 hashfull 959 tbhits 0 time 5925 pv d5c7 d8c7 a1d1 B@c3 N@d6 e8f8 B@e5 c3d2 f3d2 P@f6 P@e7 c7e7 d6c8

@ddugovic Sorry, I do not get it. Why is this a hint for a bug? The transposition seems far from forced, and can occur earliest at the 4th ply, so I think we can not make many assumptions about the relation of the scores for the two moves.

Sorry @ianfab , I'm just grasping at straws trying to understand how Rd1 is not the PV move, especially since after Rd1 is played SF's evaluation jumps: https://github.com/ddugovic/Stockfish/issues/158#issuecomment-452755029

Hey, I agree with @ianfab that the transposition seems really unlikely to occur because if first Rd1 is played then taking the knight is almost forced by black. Consequently a later Nxc7 becomes impossible.

I'm not a SF expert, but for me it seems more like a potential minor issue in move ordering or soft pruning of some sort. From a purely objective point of view without looking at the position Nxc7 is easily the best move because it comes with check and white gets at least a pawn while opening a crucial back rank square for later attack. However, it also helps black developing a counter-attack with 10. B@c3 to follow which is a lot harder to evaluate correctly.

Nevertheless, the move Nxc7 doesn't seem like a bad move to me and it's not 100% certain that Rd1 is a better move in this position. The line 9. Nxc7 Qxc7 10. Rd1 B@c3 11. N@d6 Kf8 12. Nc4 is also promising for white.

SF is a bit too materialistic at times, and in this case might slightly misevaluate the attacking chances. To me it seems like improving evaluation of threats and weak squares, which we mostly neglected for crazyhouse so far, might help to mitigate this.

That makes sense. FYI on current master now Rd1 is the PV after 7 seconds (previously 17 seconds):

info depth 1 seldepth 2 multipv 1 score cp 495 nodes 452 nps 226000 tbhits 0 time 2 pv f1b5 P@c6
info depth 2 seldepth 3 multipv 1 score cp 666 nodes 714 nps 238000 tbhits 0 time 3 pv a1b1 d8d5 B@b5 d5b5 f1b5 P@c6 b1b2
info depth 3 seldepth 4 multipv 1 score cp 365 nodes 963 nps 321000 tbhits 0 time 3 pv a1b1 d8d5 b1b2
info depth 4 seldepth 4 multipv 1 score cp 354 nodes 1254 nps 418000 tbhits 0 time 3 pv d5c7 d8c7 a1b1 B@c6 b1b2
info depth 5 seldepth 5 multipv 1 score cp 354 nodes 1420 nps 355000 tbhits 0 time 4 pv d5c7 d8c7 a1b1 B@c6 b1b2
info depth 6 seldepth 6 multipv 1 score cp 308 nodes 1784 nps 446000 tbhits 0 time 4 pv d5c7 d8c7 a1b1 b8d7 b1b2 B@c6
info depth 7 seldepth 7 multipv 1 score cp 153 nodes 3079 nps 615800 tbhits 0 time 5 pv d5c7 d8c7 a1b1 N@e4 b1b2 e4d2 f3d2
info depth 8 seldepth 8 multipv 1 score cp 544 nodes 5491 nps 686375 tbhits 0 time 8 pv a1b1 d8d5 N@c3
info depth 9 seldepth 14 multipv 1 score cp 229 nodes 17503 nps 583433 tbhits 0 time 30 pv a1b1 e6d5 B@g5 f7f6 P@f7 e8f7 N@e5 f7e7 b1b2
info depth 10 seldepth 13 multipv 1 score cp 386 nodes 28134 nps 611608 tbhits 0 time 46 pv d5c7 d8c7 a1b1 N@e4 B@g3 e4d2 f3d2
info depth 11 seldepth 17 multipv 1 score cp 66 nodes 41342 nps 617044 tbhits 0 time 67 pv d5c7 d8c7 a1b1 B@c3 N@d6 e8f8 P@f6 c3d2 f3d2 f8g8 P@e7 g7f6 e7e8q g8g7
info depth 12 seldepth 19 multipv 1 score cp 454 nodes 51279 nps 617819 tbhits 0 time 83 pv d5c7 d8c7 a1b1 N@e4 B@e5 e4d2 e5c7 d2f3 g2f3 e8g8 Q@a3
info depth 13 seldepth 22 multipv 1 score cp 83 nodes 208455 nps 599008 tbhits 0 time 348 pv d5c7 d8c7 a1d1 B@c3 B@c5 c3d2 f3d2 b8c6 P@d6 c7a5 N@c7 a5c7 d6c7 Q@c1 B@b5 c1d1 e1d1 R@c1 d1e2 b2b1q
info depth 14 seldepth 27 multipv 1 score cp 140 nodes 275526 nps 593806 tbhits 0 time 464 pv d5c7 d8c7 a1d1 B@f4 P@e3 N@e4 e3f4
info depth 15 seldepth 20 multipv 1 score cp 359 nodes 308635 nps 599291 tbhits 0 time 515 pv d5c7 d8c7 a1d1 B@f4 P@e3 N@e4 N@b5 e4d2
info depth 16 seldepth 35 multipv 1 score cp 252 nodes 802042 nps 603947 hashfull 406 tbhits 0 time 1328 pv d5c7 d8c7 a1d1 B@a5 N@d6 c7d6 d2a5 b8c6 B@c5 d6d7 B@a4 N@e4 a4c6 b7c6 P@d2 N@d6 c5d6 d7d6
info depth 17 seldepth 27 multipv 1 score cp 313 nodes 1077828 nps 601131 hashfull 535 tbhits 0 time 1793 pv d5c7 d8c7 a1d1 B@a5 P@b4 P@c3 d2g5 a5b4 f1d3 f7f6 g5h5 e8f8 e1g1 P@d2 f3d2 c3d2 d1d2
info depth 18 seldepth 31 multipv 1 score cp 160 upperbound nodes 1649324 nps 604812 hashfull 722 tbhits 0 time 2727 pv d5c7 d8c7
info depth 18 seldepth 36 multipv 1 score cp 99 upperbound nodes 2135271 nps 603354 hashfull 822 tbhits 0 time 3539 pv d5c7 d8c7
info depth 18 seldepth 36 multipv 1 score cp 148 lowerbound nodes 3802057 nps 601972 hashfull 971 tbhits 0 time 6316 pv a1d1
info depth 18 seldepth 36 multipv 1 score cp 245 nodes 4477638 nps 600944 hashfull 986 tbhits 0 time 7451 pv a1d1 e6d5 P@h6 d8e7 B@e5 N@e4 h6g7 h8g8 P@f6 e7d8 N@h6 e4d2 f3d2 B@b4 h6g8 Q@a5 g8h6 b4d2 d1d2 b2b1q N@d1 a5d2 e1d2

be5a2f015e45886e32867b4559ef51dd694a3cec fixed several search problems, so improving evaluation of threats and weak squares may further improve this:

info depth 1 seldepth 2 multipv 1 score cp 522 nodes 1030 nps 343333 tbhits 0 time 3 pv f1b5 P@c6
info depth 2 seldepth 3 multipv 1 score cp 371 nodes 1637 nps 409250 tbhits 0 time 4 pv a1b1 e6d5 b1b2
info depth 3 seldepth 3 multipv 1 score cp 349 nodes 1953 nps 390600 tbhits 0 time 5 pv d5c7 d8c7 a1b1
info depth 4 seldepth 4 multipv 1 score cp 411 nodes 2115 nps 423000 tbhits 0 time 5 pv d5c7 d8c7 a1b1
info depth 5 seldepth 5 multipv 1 score cp 411 nodes 2340 nps 468000 tbhits 0 time 5 pv d5c7 d8c7 a1b1 B@c6 b1b2
info depth 6 seldepth 6 multipv 1 score cp 515 nodes 2521 nps 504200 tbhits 0 time 5 pv d5c7 d8c7 a1b1 B@c6
info depth 7 seldepth 8 multipv 1 score cp 323 nodes 4411 nps 551375 tbhits 0 time 8 pv d5c7 d8c7 a1b1 B@c6 b1b2 b7b6 b2b3 h7h6
info depth 8 seldepth 12 multipv 1 score cp 276 nodes 11693 nps 615421 tbhits 0 time 19 pv a1d1 d8d5 P@h6 d5e4 N@e3 g7h6 B@g7
info depth 9 seldepth 11 multipv 1 score cp 353 nodes 15926 nps 612538 tbhits 0 time 26 pv a1d1 d8d5 P@h6 d5e4 N@e3 b2b1q d1b1
info depth 10 seldepth 13 multipv 1 score cp 489 nodes 18643 nps 621433 tbhits 0 time 30 pv a1d1 d8d5 P@h6 d5e4 N@e3 b2b1q
info depth 11 seldepth 15 multipv 1 score cp 244 nodes 49412 nps 633487 tbhits 0 time 78 pv a1d1 e6d5 B@g5 B@f6 N@h5 d8e7 P@e5 f6g5 f3g5 e8d8 B@f3
info depth 12 seldepth 19 multipv 1 score cp 145 nodes 122437 nps 631118 tbhits 0 time 194 pv a1d1 e6d5 P@h6 d8e7 f1e2 B@b4 h6g7 b4d2 f3d2 Q@f6 g7h8q f6h8 P@e5 b8c6
info depth 13 seldepth 19 multipv 1 score cp 292 nodes 207417 nps 628536 tbhits 0 time 330 pv d5c7 d8c7 a1b1 B@c3 N@d6 c7d6 d2c3 e8g8 c3b2 P@e4 P@e5
info depth 14 seldepth 24 multipv 1 score cp 220 nodes 429678 nps 635618 tbhits 0 time 676 pv a1d1 d8d5 P@h6 B@c1 d1c1 b2c1q d2c1 N@e4 h6g7 h8g8 B@g5 P@f6 B@b4 f6g5
info depth 15 seldepth 27 multipv 1 score cp 211 nodes 776171 nps 641463 hashfull 363 tbhits 0 time 1210 pv a1d1 e6d5 P@h6 B@e7 h6g7 h8g8 B@f8 b8c6 f1b5 e7f8 b5c6 b7c6 g7f8q e8f8 B@h6 P@g7 f3e5 g7h6
info depth 16 seldepth 35 multipv 1 score cp 208 nodes 1480463 nps 637580 hashfull 630 tbhits 0 time 2322 pv a1d1 e6d5 P@h6 d8e7 B@e5 P@f6 h6g7 h8g8 N@h5 N@e4 f1d3 e4d2 h5f6 e7f6 e5f6 d2f3 g2f3 b8c6 e1f1 B@e7 f6e7 c6e7
info depth 17 seldepth 25 multipv 1 score cp 188 upperbound nodes 1622410 nps 636488 hashfull 672 tbhits 0 time 2549 pv a1d1 e6d5
info depth 17 seldepth 26 multipv 1 score cp 173 upperbound nodes 1738341 nps 636521 hashfull 702 tbhits 0 time 2731 pv a1d1 e6d5
info depth 17 seldepth 28 multipv 1 score cp 152 upperbound nodes 1794410 nps 636993 hashfull 715 tbhits 0 time 2817 pv a1d1 e6d5
info depth 17 seldepth 28 multipv 1 score cp 169 lowerbound nodes 2033375 nps 634240 hashfull 777 tbhits 0 time 3206 pv a1d1
info depth 17 seldepth 28 multipv 1 score cp 207 lowerbound nodes 2116992 nps 634209 hashfull 793 tbhits 0 time 3338 pv a1d1
info depth 17 seldepth 28 multipv 1 score cp 151 nodes 2173775 nps 634123 hashfull 805 tbhits 0 time 3428 pv a1d1 e6d5 P@h6 B@c1 f1b5 c8d7 d2e2 N@e4 h6g7 h8g8 B@h4 P@e7 e1g1 g8g7 N@h5 g7g8 f3e5 b8c6 b5c6 b7c6 d1c1 b2c1q
info depth 18 seldepth 28 multipv 1 score cp 161 lowerbound nodes 2464823 nps 632654 hashfull 856 tbhits 0 time 3896 pv a1d1
info depth 18 seldepth 28 multipv 1 score cp 140 upperbound nodes 2596862 nps 632917 hashfull 876 tbhits 0 time 4103 pv a1d1 e6d5
info depth 18 seldepth 33 multipv 1 score cp 156 lowerbound nodes 2700050 nps 633071 hashfull 888 tbhits 0 time 4265 pv a1d1
info depth 18 seldepth 33 multipv 1 score cp 177 lowerbound nodes 2802325 nps 633580 hashfull 897 tbhits 0 time 4423 pv a1d1
info depth 18 seldepth 33 multipv 1 score cp 206 lowerbound nodes 2865723 nps 634711 hashfull 899 tbhits 0 time 4515 pv a1d1
info depth 18 seldepth 33 multipv 1 score cp 244 lowerbound nodes 2936675 nps 635781 hashfull 903 tbhits 0 time 4619 pv a1d1
info depth 18 seldepth 33 multipv 1 score cp 217 nodes 3001049 nps 635815 hashfull 909 tbhits 0 time 4720 pv a1d1 e6d5 P@h6 d8e7 B@e5 N@e4 h6g7 h8g8 N@h6 g8g7 e5g7 e4d2 f3e5 b2b1q d1b1 B@f8 e1d2 f8g7 R@g8 B@f8 h6f7
info depth 19 seldepth 24 multipv 1 score cp 205 upperbound nodes 3076791 nps 635831 hashfull 917 tbhits 0 time 4839 pv a1d1 e6d5
info depth 19 seldepth 31 multipv 1 score cp 217 lowerbound nodes 3279371 nps 635783 hashfull 934 tbhits 0 time 5158 pv a1d1
info depth 19 seldepth 31 multipv 1 score cp 232 lowerbound nodes 3451536 nps 635876 hashfull 945 tbhits 0 time 5428 pv a1d1
info depth 19 seldepth 32 multipv 1 score cp 196 upperbound nodes 3561427 nps 635175 hashfull 951 tbhits 0 time 5607 pv a1d1 e6d5
info depth 19 seldepth 32 multipv 1 score cp 224 lowerbound nodes 3844620 nps 634111 hashfull 964 tbhits 0 time 6063 pv a1d1
info depth 19 seldepth 35 multipv 1 score cp 262 lowerbound nodes 4080775 nps 634448 hashfull 972 tbhits 0 time 6432 pv a1d1
info depth 19 seldepth 35 multipv 1 score cp 312 lowerbound nodes 4423938 nps 633983 hashfull 980 tbhits 0 time 6978 pv a1d1
info depth 19 seldepth 35 multipv 1 score cp 167 upperbound nodes 4642342 nps 633507 hashfull 984 tbhits 0 time 7328 pv a1d1 e6d5
info depth 19 seldepth 35 multipv 1 score cp 272 lowerbound nodes 5057359 nps 631775 hashfull 989 tbhits 0 time 8005 pv a1d1
info depth 19 seldepth 35 multipv 1 score cp 250 nodes 5600883 nps 629312 hashfull 994 tbhits 0 time 8900 pv a1d1 e6d5 P@h6 d8e7 B@e5 N@e4 h6g7 h8g8 d2h6 b8d7 h6h7 e4f6 h7h8 B@h7 P@e6 d7e5 d4e5 h7c2 e6f7 e7f7 N@h6 c2d1 h6f7 e8f7 e5f6 b2b1q
info depth 20 seldepth 33 multipv 1 score cp 261 lowerbound nodes 6455056 nps 627252 hashfull 999 tbhits 0 time 10291 pv a1d1
info depth 20 seldepth 38 multipv 1 score cp 271 lowerbound nodes 6711751 nps 627442 hashfull 999 tbhits 0 time 10697 pv a1d1
info depth 20 seldepth 38 multipv 1 score cp 287 lowerbound nodes 6964846 nps 627746 hashfull 999 tbhits 0 time 11095 pv a1d1
info depth 20 seldepth 38 multipv 1 score cp 239 upperbound nodes 7073556 nps 627199 hashfull 999 tbhits 0 time 11278 pv a1d1 e6d5
info depth 20 seldepth 38 multipv 1 score cp 210 upperbound nodes 7890789 nps 625211 hashfull 1000 tbhits 0 time 12621 pv a1d1 e6d5
info depth 20 seldepth 38 multipv 1 score cp 242 lowerbound nodes 8584694 nps 625159 hashfull 1000 tbhits 0 time 13732 pv d5c7
info depth 20 seldepth 38 multipv 1 score cp 292 lowerbound nodes 8780908 nps 625242 hashfull 1000 tbhits 0 time 14044 pv d5c7
info depth 20 seldepth 38 multipv 1 score cp 357 lowerbound nodes 9359552 nps 625554 hashfull 1000 tbhits 0 time 14962 pv d5c7
info depth 20 seldepth 38 multipv 1 score cp 171 upperbound nodes 9526002 nps 625846 hashfull 1000 tbhits 0 time 15221 pv d5c7 d8c7
info depth 20 seldepth 44 multipv 1 score cp 306 lowerbound nodes 11682292 nps 625156 hashfull 1000 tbhits 0 time 18687 pv a1d1
info depth 20 seldepth 44 multipv 1 score cp 361 nodes 14906703 nps 621604 hashfull 1000 tbhits 0 time 23981 pv a1d1 e6d5 P@h6 N@e4 h6g7 h8g8 d2h6 c8f5 f1d3 b8c6 d3e4 f5e4 N@f6 d8f6 h6f6 B@b4 B@d2 B@e7 Q@f8 e7f8 g7f8q g8f8 d2b4 c6b4 B@a4 P@c6
info depth 21 seldepth 33 multipv 1 score cp 350 upperbound nodes 15353198 nps 621712 hashfull 1000 tbhits 0 time 24695 pv a1d1 e6d5
info depth 21 seldepth 33 multipv 1 score cp 338 upperbound nodes 15417004 nps 621979 hashfull 1000 tbhits 0 time 24787 pv a1d1 e6d5
info depth 21 seldepth 33 multipv 1 score cp 322 upperbound nodes 15601523 nps 622194 hashfull 1000 tbhits 0 time 25075 pv a1d1 e6d5
info depth 21 seldepth 33 multipv 1 score cp 300 upperbound nodes 15830058 nps 622446 hashfull 1000 tbhits 0 time 25432 pv a1d1 e6d5
info depth 21 seldepth 33 multipv 1 score cp 269 upperbound nodes 16092749 nps 623025 hashfull 1000 tbhits 0 time 25830 pv a1d1 e6d5
info depth 21 seldepth 34 multipv 1 score cp 293 lowerbound nodes 17353856 nps 621645 hashfull 1000 tbhits 0 time 27916 pv a1d1
info depth 21 seldepth 34 multipv 1 score cp 244 nodes 17891337 nps 621788 hashfull 1000 tbhits 0 time 28774 pv a1d1 e6d5 P@h6 N@e4 h6g7 h8g8 d2h6 c8f5 f1d3 b8d7 e1g1 P@h3 P@e6 f7e6 d3e4 d5e4 B@h5 P@g6 f3e5 B@f7 e5f7 d8e7 B@a4 h3g2 a4d7 e8d7 N@e5 d7c8 h6h7 g2f1q d1f1

I'm deferring further investigation until after official-stockfish/Stockfish#2557 merges (since it sounds quite useful).

I have tried various evaluation changes pertaining to threats and weak squares.

IM SpinalTap says that in chess, 4 pieces are necessary for an attack; in shogi the same rule applies. Our evaluation function is quadratic, not quarternary; but perhaps making it at least ternary (considering combinations of 3 pieces at a time) may help with measuring threats and types of weaknesses with respect to piece combinations? I don't know.

I am closing this as a possible duplicate of #94 .

Was this page helpful?
0 / 5 - 0 ratings

Related issues

Vinvin20 picture Vinvin20  路  14Comments

niklasf picture niklasf  路  10Comments

niklasf picture niklasf  路  5Comments

ianfab picture ianfab  路  6Comments

Vinvin20 picture Vinvin20  路  12Comments