Stockfish Testing Queue

Finished - 1408 tests

19-05-03 31m tweak_KFA2^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 58889 W: 13222 L: 13152 D: 32515
sprt @ 10+0.1 th 1 Built upon @Vizvezdenec's 70K STC green / 126K LTC yellow, a combo of tweaks by @SFisGOD and myself. Introduce compensatory tweak in the kingDanger constant term. -14.
19-05-03 31m tweak_KFA2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 59749 W: 13386 L: 13312 D: 33051
sprt @ 10+0.1 th 1 Opposite effect: kingDanger constant 0.
19-04-29 31m tweak_flankAttacks diff
LLR: -1.55 (-2.94,2.94) [0.00,4.00]
Total: 119434 W: 26605 L: 26218 D: 66611
sprt @ 10+0.1 th 1 After 80K-and-counting STC games, both MG -= 1 and EG -= 1 tests are ongoing and appear strong (+1.46 Elo, +1.15 Elo). Try both together.
19-04-30 31m tweak_flankAttacks diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 22575 W: 3774 L: 3855 D: 14946
sprt @ 60+0.6 th 1 LLR is near 0 after 111K STC games. Since that test isn't likely to finish soon, speculative LTC. Low throughput.
19-04-29 31m tweak_flankAttacks diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 111303 W: 24788 L: 24511 D: 62004
sprt @ 10+0.1 th 1 EG -= 1
19-04-29 31m tweak_flankAttacks^^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 103889 W: 23099 L: 22852 D: 57938
sprt @ 10+0.1 th 1 MG -= 1
19-04-29 31m tweak_flankAttacks^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 19887 W: 4376 L: 4462 D: 11049
sprt @ 10+0.1 th 1 EG += 1
19-04-29 31m tweak_flankAttacks^^^ diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 15707 W: 3433 L: 3535 D: 8739
sprt @ 10+0.1 th 1 Small tweaks to FlankAttacks on a fairly empty framework. MG += 1
19-04-28 31m passers_3connected^^^ diff
LLR: -2.96 (-2.94,2.94) [0.00,3.50]
Total: 25740 W: 4305 L: 4409 D: 17026
sprt @ 60+0.6 th 1 Speculative LTC for 65K yellow, since connected, advanced passers are mostly an endgame feature, and endgames are more important at LTC. Low throughput.
19-04-28 31m passers_3connected diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 65450 W: 14645 L: 14484 D: 36321
sprt @ 10+0.1 th 1 Require that the pawn be a "true" passer (i.e., not a candidate passer). Remove more_than_one. +50%
19-04-28 31m passers_3connected diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 21419 W: 4719 L: 4777 D: 11923
sprt @ 10+0.1 th 1 +75%
19-04-28 31m passers_3connected diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 24911 W: 5597 L: 5637 D: 13677
sprt @ 10+0.1 th 1 Effect size tweaks of my best version so far. Require that the pawn be a "true" passer (i.e., not a candidate passer). Double effect: +100%.
19-04-28 31m passers_3connected^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 7760 W: 1691 L: 1818 D: 4251
sprt @ 10+0.1 th 1 Effect size tweaks of my best version so far. Require that the pawn be a "true" passer (i.e., not a candidate passer). Half effect: +25%.
19-04-28 31m passers_3connected diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 11981 W: 2635 L: 2740 D: 6606
sprt @ 10+0.1 th 1 Restore more_than_one and double effect.
19-04-27 31m passers_3connected diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 32373 W: 7251 L: 7254 D: 17868
sprt @ 10+0.1 th 1 We have seen a pattern of static eval underestimating advanced neighboring passers (most recently in Bryan's analysis of CCC7 Blitz Bonanza Final Game 353). I wonder if previous attempts have failed because they augmented k rather than w: the latter is rank-dependent, and clearly the danger of connected passers increases with rank. Start with large effect: w += 50% if there are multiple nearby passers.
19-04-27 31m passers_3connected diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 24717 W: 5399 L: 5442 D: 13876
sprt @ 10+0.1 th 1 Broader effect (remove more_than_one), but still large effect size.
19-04-27 31m tweak_KDC diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 26566 W: 5845 L: 5904 D: 14817
sprt @ 10+0.1 th 1 Double effect. Test on top of PR #2118.
19-04-27 31m tweak_KDC diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 25607 W: 5649 L: 5712 D: 14246
sprt @ 10+0.1 th 1 @Vizvezdenec's Elo-gaining PR #2118 appears to increase the average kingDanger (over bench positions) by about 72 units. Therefore, experiment with further raising the kingDanger threshold. Test on top of PR #2118. (Fixed bench: I forgot that fishtest doesn't automatically load the base bench from the commit message.)
19-04-24 31m extend_2check diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 25797 W: 5645 L: 5682 D: 14470
sprt @ 10+0.1 th 1 Double extension for discovered checks involving an advanced pawn push.
19-04-24 31m extend_discCheck diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 30167 W: 6669 L: 6684 D: 16814
sprt @ 10+0.1 th 1 Narrow the discovered check extension to not include discovered checks that land beside the enemy king (which then may be able to capture).
19-04-24 31m extend_rookconnection diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 19873 W: 4349 L: 4415 D: 11109
sprt @ 10+0.1 th 1 Exclude king moves.
19-04-24 31m extend_rookconnection diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 22682 W: 4961 L: 5013 D: 12708
sprt @ 10+0.1 th 1 Extend moves that unblock/connect our rooks along the back rank.
19-04-24 31m extend_rookconnection diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 12963 W: 2826 L: 2926 D: 7211
sprt @ 10+0.1 th 1 Extend killer moves that connect our rooks.
19-04-24 31m tweak_Bmob diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 12352 W: 2684 L: 2800 D: 6868
sprt @ 10+0.1 th 1 Replace the BishopPawns tweak with the best-performing of the four I tested, S(2, 7).
19-04-24 31m tweak_Bmob^ diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 15227 W: 3356 L: 3460 D: 8411
sprt @ 10+0.1 th 1 Something about @xoroshiro's Bmob seemed strange to me--I had just tested the exact same tweak to BishopPawns as in that test, and it was -1.5 Elo. Try testing the remainder of @xoroshiro's raw values without this one suspicious tweak.
19-04-24 31m extend_pawnlock diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 29648 W: 6603 L: 6620 D: 16425
sprt @ 10+0.1 th 1 Same as last take, but with NPM condition restored.
19-04-23 31m extend_pawnlock diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 14391 W: 3146 L: 3239 D: 8006
sprt @ 10+0.1 th 1 Extend pawn moves such that we are immediately stopped by an enemy pawn, and neither we nor the opponent have pawn breaks available (i.e., there are no pawns on adjacent files which can be pushed to attack).
19-04-23 31m extend_pawnlock diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 16664 W: 3637 L: 3719 D: 9308
sprt @ 10+0.1 th 1 Add NPM condition.
19-04-23 31m extend_pawnlock diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 7344 W: 1510 L: 1638 D: 4196
sprt @ 10+0.1 th 1 Extension for a pawn move that locks up the pawn structure: immediately stopped by enemy pawn and no enemy pawns on adjacent files to attack it. (Fix inconsistent bench: exclude promotions. Trying to find the blocking square above to_sq when to_sq is on RANK_8 created problems.)
19-04-23 31m extend_castleOpp diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 24104 W: 5389 L: 5434 D: 13281
sprt @ 10+0.1 th 1 Restrict castling extension to opposite-sides castling.
19-04-23 31m killerApp_check diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 17203 W: 3782 L: 3861 D: 9560
sprt @ 10+0.1 th 1 Add NPM condition.
19-04-23 31m combo_BP^ diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 33127 W: 7304 L: 7337 D: 18486
sprt @ 10+0.1 th 1 Combo with @SFisGOD's PassedRank_twk (March 7, LTC 118K yellow).
19-04-23 31m combo_BP diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 20866 W: 4566 L: 4648 D: 11652
sprt @ 10+0.1 th 1 Combo with @SFisGOD's backward_twkr (March 9, LTC 60K yellow).
19-04-23 31m tweak_KFA diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 14284 W: 3131 L: 3239 D: 7914
sprt @ 10+0.1 th 1 Revive this tweak by @SFisGOD. It was a LTC 70K yellow on March 10, but probably interacts with the tweaks I made to the kingDanger constant and cutoff on April 9. Does this change the performance of this tweak?
19-04-22 31m tweak_BishopPawns^^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 57877 W: 12872 L: 12807 D: 32198
sprt @ 10+0.1 th 1 MG -= 1.
19-04-22 31m extend_pawnPushVsK3 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 17849 W: 3943 L: 4019 D: 9887
sprt @ 10+0.1 th 1 Another take on this idea.
19-04-22 31m killerApp_check diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 11955 W: 2616 L: 2721 D: 6618
sprt @ 10+0.1 th 1 We recently added an extension for killer advanced pawn pushes for passed pawns. Also extend killer advanced pawn pushes that give check.
19-04-22 31m tweak_BishopPawns^^^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 22114 W: 4923 L: 5000 D: 12191
sprt @ 10+0.1 th 1 evaluate.cpp has some small, very Elo-sensitive parameters which can only be tuned effectively by repeated tests (they're too small for SPSA). Since this hasn't been done in a while, and the framework is spending considerable time empty or partially empty, this seems like a potential opportunity for a bit of easy Elo. Start with BishopPawns. MG += 1.
19-04-22 31m tweak_BishopPawns^ diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 17195 W: 3801 L: 3897 D: 9497
sprt @ 10+0.1 th 1 EG += 1.
19-04-22 31m tweak_BishopPawns diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 15091 W: 3353 L: 3458 D: 8280
sprt @ 10+0.1 th 1 EG -= 1.
19-04-22 31m Outpost_pawnpush^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 16971 W: 3657 L: 3738 D: 9576
sprt @ 10+0.1 th 1 Currently-occupied case only, double effect.
19-04-22 31m Outpost_pawnpush diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 10454 W: 2304 L: 2417 D: 5733
sprt @ 10+0.1 th 1 Currently-occupied case only, half effect.
19-04-22 31m Outpost_pawnpush^ diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 29493 W: 6529 L: 6547 D: 16417
sprt @ 10+0.1 th 1 Only the currently-occupied case.
19-04-22 31m Outpost_pawnpush diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 12973 W: 2815 L: 2915 D: 7243
sprt @ 10+0.1 th 1 Only the future-outpost case.
19-04-22 31m Outpost_pawnpush^^ diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 8398 W: 1806 L: 1929 D: 4663
sprt @ 10+0.1 th 1 Currently, we double an Outpost bonus if the outpost is protected by a pawn. Additionally, +50% Outpost if we can protect it with a pawn push.
19-04-22 31m simplify_ThreatBy diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 10178 W: 2127 L: 2311 D: 5740
sprt @ 10+0.1 th 1 Does this branch actually produce a worthwhile speedup? Remove it. Non-functional change. (Add "?w=1", no quotes, to the diff's URL to ignore whitespace changes, which are most of the changes in this patch.)
19-04-22 31m kingDanger_2N diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 24563 W: 5403 L: 5446 D: 13714
sprt @ 10+0.1 th 1 +50 kingDanger if the opponent has 2 knights in our king flank but we have none.
19-04-22 31m BishopPawns_forward diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 6908 W: 1529 L: 1660 D: 3719
sprt @ 10+0.1 th 1 Try modifying the other term: exclude pawns behind the bishop.
19-04-21 31m BishopPawns_forward diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 8542 W: 1847 L: 1970 D: 4725
sprt @ 10+0.1 th 1 Higher weight for pawns in front of bishop.
19-04-21 31m BishopPawns_forward diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 5153 W: 1085 L: 1224 D: 2844
sprt @ 10+0.1 th 1 For the BishopPawns penalty multiplier, only consider pawns in front of the bishop, not behind or beside it.