Stockfish Testing Queue

Finished - 1408 tests

19-04-21 31m kingDanger_2N diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 14166 W: 3094 L: 3188 D: 7884
sprt @ 10+0.1 th 1 + 100.
19-04-21 31m kingDanger_2N^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 6183 W: 1335 L: 1470 D: 3378
sprt @ 10+0.1 th 1 + 25.
19-04-21 31m kingDanger_2N diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 31347 W: 6932 L: 6941 D: 17474
sprt @ 10+0.1 th 1 Idea based on a post from Jesse M on the forum. Increased kingDanger penalty if the enemy has two knights in our king flank. + 50.
19-04-21 31m 842692d5b906984b340beb9 diff
LLR: -2.95 (-2.94,2.94) [0.00,3.50]
Total: 14541 W: 2325 L: 2460 D: 9756
sprt @ 60+0.6 th 1 With more than half of our workers idle, speculative LTC to check scaling of the best of the many yellow tests on the KD_relrank branch. STC 61K yellow. Low throughput.
19-04-21 31m extend_passercreation2 diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 12150 W: 2660 L: 2764 D: 6726
sprt @ 10+0.1 th 1 Narrow tweak of 54K yellow (include one additional rank), in case its improved performance was not merely chance.
19-04-21 31m extend_passercreation2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 16125 W: 3567 L: 3652 D: 8906
sprt @ 10+0.1 th 1 In my previous attempt to only include not-pawn-defended stoppers, there was a bug. Try again.
19-04-21 31m extend_passercreation2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 54014 W: 11932 L: 11830 D: 30252
sprt @ 10+0.1 th 1 Require that the would-be passer be at least somewhat advanced.
19-04-21 31m extend_passercreation2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 22232 W: 4866 L: 4921 D: 12445
sprt @ 10+0.1 th 1 I overlooked some candidate passed pawns in my first tests. Include them here, broadening the effect. (Change "else if" to "if")
19-04-21 31m extend_passercreation2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 11410 W: 2494 L: 2602 D: 6314
sprt @ 10+0.1 th 1 Include stoppers of even more candidate passed pawns. (Also simpler code-wise.)
19-04-21 31m extend_passercreation2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 16101 W: 3546 L: 3631 D: 8924
sprt @ 10+0.1 th 1 Possibly a more precise definition: exclude pawn-defended stoppers. (In that case, the enemy defender recaptures, and our pawn is still not passed.)
19-04-21 31m extend_passercreation2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 25201 W: 5595 L: 5634 D: 13972
sprt @ 10+0.1 th 1 Building upon the recent killer passed pawn extension: extend killer captures that create passed pawns.
19-04-21 31m KD_relrank diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 21818 W: 4801 L: 4857 D: 12160
sprt @ 10+0.1 th 1 This has produced many promising yellows, but I'm running out of ideas to make it green. Coefficient 2, MG-only quadratic. (Half effect of 61K yellow.)
19-04-21 31m extend_passercreation2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 4549 W: 951 L: 1094 D: 2504
sprt @ 10+0.1 th 1 Sanity check: no killer condition.
19-04-21 31m KD_relrank diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 26427 W: 5817 L: 5851 D: 14759
sprt @ 10+0.1 th 1 Best version (quadratic, middlegame only, coefficient 4) with compensation in the KD constant term.
19-04-21 31m KD_relrank diff
LLR: -2.94 (-2.94,2.94) [0.50,4.50]
Total: 13748 W: 3065 L: 3161 D: 7522
sprt @ 10+0.1 th 1 Middlegame-only quadratic, double coefficient (8).
19-04-20 31m KD_relrank diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 61223 W: 13671 L: 13531 D: 34021
sprt @ 10+0.1 th 1 Quadratic effect (coefficient 4), middlegame only.
19-04-20 31m KD_relrank diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 38483 W: 8582 L: 8555 D: 21346
sprt @ 10+0.1 th 1 Effect size 12.
19-04-20 31m KD_relrank diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 26142 W: 5740 L: 5775 D: 14627
sprt @ 10+0.1 th 1 Confusingly, this is a consistent small gain no matter how many times I double the effect size (i.e., for 2, 4, and 8). Keep going: effect size 16.
19-04-20 31m KD_relrank diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 40164 W: 8875 L: 8841 D: 22448
sprt @ 10+0.1 th 1 Effect size 8.
19-04-20 31m KD_relrank diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 39077 W: 8692 L: 8663 D: 21722
sprt @ 10+0.1 th 1 Effect size 4.
19-04-20 31m KD_relrank diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 54747 W: 12288 L: 12180 D: 30279
sprt @ 10+0.1 th 1 Double effect.
19-04-20 31m cpasser_supporting2^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 17378 W: 3797 L: 3876 D: 9705
sprt @ 10+0.1 th 1 Only if there are multiple sub-candidate passers, e.g., pawns that would be candidate passers, but do not outnumber leverPush with their phalanx pawns, only with their phalanx and supporting pawns combined.
19-04-20 31m cpasser_supporting2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 10503 W: 2260 L: 2373 D: 5870
sprt @ 10+0.1 th 1 Bonus divisor 3.
19-04-20 31m KD_relrank^ diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 9874 W: 2133 L: 2249 D: 5492
sprt @ 10+0.1 th 1 Simply increase kingDanger as the relative rank of our king increases.
19-04-20 31m tweak_outpost2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 19193 W: 4260 L: 4348 D: 10585
sprt @ 10+0.1 th 1 In December this was highly promising (STC: 190K yellow; LTC: 66K yellow). Respin to evaluate whether it has maintained its performance.
19-04-20 31m cpasser_supporting2 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 42830 W: 9443 L: 9396 D: 23991
sprt @ 10+0.1 th 1 Try again with larger divisor (4) and speedup.
19-04-20 31m cpasser_supporting2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 7565 W: 1647 L: 1775 D: 4143
sprt @ 10+0.1 th 1 Revisiting this idea. Store these less-than-candidate-passers separately and give bonus for them, but less than an ordinary candidate passer.
19-04-19 31m 670c970a3b923cbe4513050 diff
LLR: -2.95 (-2.94,2.94) [0.00,3.50]
Total: 30639 W: 5219 L: 5308 D: 20112
sprt @ 60+0.6 th 1 There's no reason to leave the framework empty. Speculative LTC for @mstembera's 125K STC yellow.
19-04-18 31m tweak_candidatePasser2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 22534 W: 4966 L: 5041 D: 12527
sprt @ 10+0.1 th 1 Very simple patch: lower candidatePasser bonuses (/ 3 rather than / 2).
19-04-12 31m extend_pawnPushVsK2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 48618 W: 10802 L: 10726 D: 27090
sprt @ 10+0.1 th 1 Restore NPM condition.
19-04-12 31m lmr_pawnPushVsK3 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 34089 W: 7596 L: 7591 D: 18902
sprt @ 10+0.1 th 1 Remove APP condition and replace with NPM.
19-04-12 31m BackwardIsolatedPawn diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 8147 W: 1689 L: 1813 D: 4645
sprt @ 10+0.1 th 1 Joining on this particular investigation by @OuaisBla and @ElbertoOne. If a pawn is both Isolated and Backward, apply the average of the two penalties.
19-04-12 31m lmr_pawnPushVsK3 diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 26515 W: 5835 L: 5868 D: 14812
sprt @ 10+0.1 th 1 Return to the LMR approach that was more promising in the past, but now using move == ss->killers[0].
19-04-12 31m extend_pawnPushVsK2 diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 23766 W: 5229 L: 5276 D: 13261
sprt @ 10+0.1 th 1 My old ideas about pawn pushes against the enemy king struggled because they were too broad, despite attempts at non-pawn-material conditions, etc. I like the recent idea of using ss->killers[0]. In addition to extending killer advanced pawn pushes that create passers, extend killer advanced pawn pushes near the enemy king.
19-04-11 31m cpasser_supporting^^ diff
LLR: -2.95 (-2.94,2.94) [0.00,3.50]
Total: 78183 W: 13350 L: 13304 D: 51529
sprt @ 60+0.6 th 1 Speculative LTC for 70K yellow. Low throughput (500).
19-04-11 31m cpasser_supporting diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 15922 W: 3480 L: 3566 D: 8876
sprt @ 10+0.1 th 1 Like 70K yellow, but with greater compensation.
19-04-10 31m cpasser_supporting^ diff
LLR: -2.95 (-2.94,2.94) [0.50,4.50]
Total: 70273 W: 15644 L: 15460 D: 39169
sprt @ 10+0.1 th 1 Adjust the bonus the opposite direction.
19-04-10 31m cpasser_supporting diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 14746 W: 3255 L: 3347 D: 8144
sprt @ 10+0.1 th 1 Restore 56K yellow, and adjust the magnitude of the candidate passer bonus to compensate for the broader effect.
19-04-10 31m cpasser_supporting diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 43912 W: 9858 L: 9804 D: 24250
sprt @ 10+0.1 th 1 Broader effect: pawns defending s, phalanx pawns, and pawns defended by s.
19-04-09 31m cpasser_supporting diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 56034 W: 12451 L: 12338 D: 31245
sprt @ 10+0.1 th 1 I believe I've isolated the reason that the pawns in SF 0-1 Antifish (Game 37 of CCC7 Blitz Bonanza Final) are not considered candidate passers (see Bryan's analysis in the forum); revise that here. We currently require that there be at least as many phalanx pawns as leverPush. Instead, also include pawns we are supporting alongside phalanx pawns.
19-04-09 31m combo_KDC^ diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 30708 W: 5296 L: 5043 D: 20369
sprt @ 60+0.6 th 1 Speculative LTC for 141K yellow. Hopefully this is the answer...
19-04-09 31m simplify_kingDangerCuto diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 10284 W: 2178 L: 2363 D: 5743
sprt @ 10+0.1 th 1 @MichaelB7's idea: remove the kingDanger threshold entirely.
19-04-08 31m combo_KDC diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 141225 W: 31239 L: 30846 D: 79140
sprt @ 10+0.1 th 1 Combo of tweaks to kingDanger cutoff and constant: try @Vizvezdenec's HandTune18 (constant -15; STC 74K yellow on March 18).
19-04-08 31m combo_KDC diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 22002 W: 4804 L: 4882 D: 12316
sprt @ 10+0.1 th 1 Combo of tweaks to kingDanger: try @Vizvezdenec's HandTune2 (STC 71K yellow, March 18).
19-04-08 31m combo_KDC^ diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 10404 W: 2228 L: 2351 D: 5825
sprt @ 10+0.1 th 1 Combo of tweaks to kingDanger cutoff and constant, where there are obviously opportunities for interactions. Try constant 30 (which I have tested recently: it appeared to be a small STC gain).
19-04-08 31m combo_ver_KDcutoff diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 29037 W: 6313 L: 6363 D: 16361
sprt @ 10+0.1 th 1 Looking for a combo to make this 161K LTC yellow green. Try @DU-jdto's verification tweak. (Do you have a recent promising [0, 4]? Please let me know in a comment!)
19-04-07 31m kingDanger_cutoff^ diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 33485 W: 5575 L: 5622 D: 22288
sprt @ 60+0.6 th 1 Surprised that 50 performed better than 150, but here's the LTC. Set throughput based on whether the STC passes. Priority -1 until STC ends.
19-04-07 31m kingDanger_cutoff^ diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 76257 W: 16830 L: 16693 D: 42734
sprt @ 10+0.1 th 1 A cutoff of 100 was a 161K LTC[0,4] yellow: LOS 94.3%, Elo = 1.06 [-0.30,2.52]. It would have passed [0, 3.5]. Try a few small tweaks and hope to find the extra 0.1 Elo this needs. Cutoff 50.
19-04-07 31m kingDanger_cutoff diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 39043 W: 8580 L: 8590 D: 21873
sprt @ 10+0.1 th 1 Cutoff 150.
19-04-04 31m kingDanger_negative2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 161519 W: 27413 L: 27046 D: 107060
sprt @ 60+0.6 th 1 Merge new master and speculative LTC (low throughput) to evaluate whether this 74K yellow may actually be a gain.