Stockfish Testing Queue

Finished - 24198 tests

01-05-15 sg bishops diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 9229 W: 1708 L: 1789 D: 5732
sprt @ 15+0.05 th 1 Inspired by fauzi's post i try it the other way around. Penalty for bishop if hindered by own blocked pawns
18-04-15 sg king_safety diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 73859 W: 14291 L: 14199 D: 45369
sprt @ 15+0.05 th 1 Retest take 2 with weight=1: Very small penalty if pawn in king ring is attacked by opponent
30-04-15 Vo Pufferfish diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18292 W: 3451 L: 3508 D: 11333
sprt @ 15+0.05 th 1 Don't update history stats for repetitive move.
28-04-15 jk see diff
ELO: -10.03 +-3.0 (95%) LOS: 0.0%
Total: 20000 W: 3534 L: 4111 D: 12355
20000 @ 15+0.05 th 1 SEE vs. MVV
29-04-15 SC MVV_MAV diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4632 W: 945 L: 827 D: 2860
sprt @ 15+0.05 th 1 On bench PieceValue - 200* relativeRank is a better approximation of pos.see than PieceValue alone (Pearson correlation from 0.55 to 0.59). See whether it is enough to beat MVV.
29-04-15 lb psqt diff
ELO: -18.59 +-7.3 (95%) LOS: 0.0%
Total: 3629 W: 655 L: 849 D: 2125
20000 @ 15+0.05 th 1 new psqt: take 2 (this time tuned in 3+0.03)
28-04-15 sn generalize_hanging diff
LLR: -3.32 (-2.94,2.94) [-1.50,4.50]
Total: 23195 W: 4376 L: 4433 D: 14386
sprt @ 15+0.05 th 1 The usual hanging penalty applies to weak pieces which have zero defenders. This patch generalize this for weak pawns by comparing attack and defense, even if defense is not zero.
29-04-15 Fi TTSmartSave diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 11474 W: 2141 L: 2216 D: 7117
sprt @ 15+0.05 th 1 Don't overwrite any existing TT data with blanks. 2MB
26-04-15 mc null_tune diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 55092 W: 10498 L: 10461 D: 34133
sprt @ 15+0.05 th 1 Tuned null search reduction: take 2
28-04-15 Vo Tilefish3 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 12996 W: 2486 L: 2557 D: 7953
sprt @ 15+0.05 th 1 Use depth*depth*log(depth) only for history. This version should be much stronger...
26-04-15 jo union_psqt diff
LLR: -3.79 (-2.94,2.94) [0.00,4.00]
Total: 54258 W: 10514 L: 10523 D: 33221
sprt @ 15+0.05 th 1 Union of knight and rook psqt values.
25-04-15 lb union diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 72364 W: 11612 L: 11455 D: 49297
sprt @ 60+0.05 th 1 that one should be strong enough to pass. if not i give up!
27-04-15 lb psqt diff
ELO: -14.16 +-3.1 (95%) LOS: 0.0%
Total: 20380 W: 3936 L: 4766 D: 11678
30000 @ 15+0.05 th 1 new psqt: 41 param
26-04-15 Ro RCC diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 43811 W: 8444 L: 8432 D: 26935
sprt @ 15+0.05 th 1 Full penalty gave 50/50 result for the new rook_support contact checks, so let's try about half penalty.
24-04-15 Ro LessSafeCheck diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 22417 W: 4313 L: 4359 D: 13745
sprt @ 15+0.05 th 1 Avoid to compute some safe check by sliders when the landing square is already a Queen safe check.
26-04-15 Fi BMCTime diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 25116 W: 3989 L: 3975 D: 17152
sprt @ 60+0.05 th 1 If a move is 100% stable from the 3rd iteration on use only 2/3 of the available time. SPSA tuned. LTC
26-04-15 Vo Tilefish2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 4979 W: 916 L: 1009 D: 3054
sprt @ 15+0.05 th 1 fixed; forgot to remove 3* factor in movepicker also sync and match against passed Rockfish.
21-04-15 SC pieceValuesMP_simple2 diff
LLR: -1.91 (-2.94,2.94) [-3.00,1.00]
Total: 50948 W: 8027 L: 8204 D: 34717
sprt @ 60+0.05 th 1 As Joona pointed out: MVV/LVA also aims to define exactly the ordering in which captures are searched, which is left open by MVV only. Try a more compact implementation of MVV/LVA. LTC.
26-04-15 Vo Tilefish2 diff
LLR: -0.36 (-2.94,2.94) [-1.50,4.50]
Total: 3240 W: 636 L: 640 D: 1964
sprt @ 15+0.05 th 1 cmhBonus=depth*depth*log(depth).
23-04-15 mc null_tune diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 57004 W: 10916 L: 10872 D: 35216
sprt @ 15+0.05 th 1 Tuned null search reduction
25-04-15 Ha pattackcount4 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 17469 W: 3355 L: 3414 D: 10700
sprt @ 15+0.05 th 1 Final try on this idea
24-04-15 Vo Rockfish-LTC diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 13212 W: 2239 L: 2045 D: 8928
sprt @ 60+0.05 th 1 h<0 and cmh<=0. Strange how the bench went from 7664232 to 8420865, when I rebased on the latest commit.
24-04-15 Fi BMCTime diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 13493 W: 2715 L: 2572 D: 8206
sprt @ 15+0.05 th 1 If a move is 100% stable from the 3rd iteration on use only 2/3 of the available time. SPSA tuned.
24-04-15 jo rook_psqt^ diff
ELO: -1.25 +-2.5 (95%) LOS: 16.4%
Total: 30000 W: 6057 L: 6165 D: 17778
30000 @ 9+0.05 th 1 Some interpolated values from tuning ranks and files separately. Take 1. (Check if one of the two tries is worth a SPRT run)
23-04-15 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 8562 W: 1607 L: 1690 D: 5265
sprt @ 15+0.05 th 1 My last try on this topic. weight=2: if pawn in king ring is attacked by opponent knight, bishop or rook. This is like a bonus for the possibility to sacrifice a piece to break the pawn shelter
24-04-15 jo rook_psqt diff
ELO: 1.30 +-2.5 (95%) LOS: 84.4%
Total: 30000 W: 6172 L: 6060 D: 17768
30000 @ 9+0.05 th 1 Some manual adjustments on the default values, hinted by tuning ranks and files separately. Take 2.
19-04-15 vi measure_pruning diff
ELO: -2.66 +-2.4 (95%) LOS: 1.5%
Total: 25472 W: 3939 L: 4134 D: 17399
40000 @ 60+0.05 th 1 Confirm that there really is still no overall Elo penalty from allowing futility pruning at PV nodes. Lower throughput. See https://groups.google.com/d/msg/fishcooking/tsV_gaWUC5A/3ZQEZJkk4TAJ
21-04-15 SC pieceValuesMP_simple diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 69976 W: 11056 L: 11011 D: 47909
sprt @ 60+0.05 th 1 Would pure MMV logic be enough? LTC
18-04-15 SC bestMoveChanges diff
ELO: -5.05 +-4.3 (95%) LOS: 1.1%
Total: 8387 W: 1377 L: 1499 D: 5511
20000 @ 60+0.2 th 1 Scales BestMoveChanges with the log2 of the moveCount to see if it makes any difference. Testing with 30+0.01 to be a multiple of TCEC time control and with nodestime since change is only in time management.
23-04-15 hx imbal diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 25146 W: 4873 L: 4759 D: 15514
sprt @ 15+0.05 th 1 imbalance vs psqt
23-04-15 lb union diff
LLR: 2.95 (-2.94,2.94) [-0.50,4.50]
Total: 23793 W: 4632 L: 4421 D: 14740
sprt @ 15+0.05 th 1 that one should be strong enough to pass. if not i give up!
23-04-15 Ha pattackcount3 diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 17299 W: 2786 L: 2807 D: 11706
sprt @ 60+0.05 th 1 Take 2: LTC
23-04-15 Vo Tilefish diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 15173 W: 2848 L: 2913 D: 9412
sprt @ 15+0.05 th 1 History Bonus=depth*depth*log(depth)
22-04-15 Fi BMCTimeTune diff
19658/35000 iterations
39426/70000 games played
70000 @ 15+0.05 th 1 Restarting tuning with higher ck because change was very slow.
22-04-15 Vo Rockfish3 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 56056 W: 10772 L: 10728 D: 34556
sprt @ 15+0.05 th 1 LMR condition: cmh <=0 && cmh+h<0
18-04-15 lb assorted diff
LLR: -2.01 (-2.94,2.94) [-0.50,4.50]
Total: 158404 W: 30625 L: 30146 D: 97633
sprt @ 15+0.05 th 1 assorted eval features tuned locally
19-04-15 SC bestMoveChanges diff
ELO: 1.18 +-2.8 (95%) LOS: 79.6%
Total: 20000 W: 3404 L: 3336 D: 13260
20000 @ 60+0.2 th 1 Since time consumption in instable moves is increased by the new scaling, decrease overall optimum time by 20%. If it improves with respect to previous verison, I will tune it.
23-04-15 Ro RookOnKingRing diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 3331 W: 631 L: 729 D: 1971
sprt @ 15+0.05 th 1 Exclude Rook from king attackers array computation.
22-04-15 aj safe_checks_fix diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18664 W: 3592 L: 3648 D: 11424
sprt @ 15+0.05 th 1 Fix my previous patch(safe_checks) where attackunits was not initialized correctly: STC
22-04-15 Vo Perch diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 9099 W: 1669 L: 1750 D: 5680
sprt @ 15+0.05 th 1 Don't update History Stats if prior move=none
22-04-15 lb pawn diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 7993 W: 1484 L: 1658 D: 4851
sprt @ 15+0.05 th 1 tuned pawn stuff: take 2
21-04-15 Ha pattackcount3 diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 46385 W: 9060 L: 8829 D: 28496
sprt @ 15+0.05 th 1 Take 2: Holy hell if this doesn't work I will stab someone
22-04-15 Ro RookOnKingRing diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 4445 W: 828 L: 923 D: 2694
sprt @ 15+0.05 th 1 Extend the King Ring left and right for the Rook attacks computation.
22-04-15 sn long_chains diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 6558 W: 1230 L: 1319 D: 4009
sprt @ 15+0.05 th 1 Take 2: S(25,25)
22-04-15 sn long_chains diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 7945 W: 1475 L: 1560 D: 4910
sprt @ 15+0.05 th 1 Bonus for pawns on the 4th and 5th ranks which are both supported by pawns and supporting other pawns. Take 1: more value for long chains in endgame.
20-04-15 SC pieceValuesMP_simple diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 1340 W: 202 L: 304 D: 834
sprt @ 15+0.05 th 1 For completeness: is pure LVA logic good enough?
21-04-15 Fi BMCTimeTune diff
9568/37500 iterations
19265/75000 games played
75000 @ 15+0.05 th 1 Tune some weights for a BestMoveChanges time patch.
19-04-15 mc null_table diff
40911/40000 iterations
63558/80000 games played
80000 @ 30+0.05 th 1 Tune null search reduction
20-04-15 jo tuned_checks diff
LLR: -2.97 (-2.94,2.94) [0.00,4.00]
Total: 48623 W: 9423 L: 9408 D: 29792
sprt @ 15+0.05 th 1 Tuned values for safe checks.
21-04-15 Ha connected_simplify diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 28948 W: 5626 L: 5839 D: 17483
sprt @ 15+0.05 th 1 Simplify connected