Stockfish Testing Queue

Finished - 35494 tests

16-10-01 sni safe_checks3 diff
LLR: -1.17 (-2.94,2.94) [-3.00,1.00]
Total: 64120 W: 8049 L: 8189 D: 47882
sprt @ 60+0.6 th 1 LTC: Simplify SafeCheck
16-10-07 Voy lmrIdea diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5341 W: 891 L: 993 D: 3457
sprt @ 10+0.1 th 1 stc
16-10-06 SC stonewallSN diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 60320 W: 7774 L: 7698 D: 44848
sprt @ 60+0.6 th 1 Try tuned stonewall boni. LTC.
16-10-07 fau try diff
ELO: -0.73 +-4.1 (95%) LOS: 36.3%
Total: 10000 W: 1790 L: 1811 D: 6399
10000 @ 10+0.1 th 1 need to know which one is the stronger one.
16-10-06 Fis pvFactor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 21882 W: 3827 L: 3860 D: 14195
sprt @ 10+0.1 th 1 If bestMoveChanges is extremely low move faster.
16-10-06 Fis pvFactor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 44627 W: 7970 L: 7907 D: 28750
sprt @ 10+0.1 th 1 Take 2 prevents consecutive easy move. (Take 1 set to pri -1 for now).
16-10-06 sni discov_checks_in_see7 diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 55029 W: 9930 L: 9574 D: 35525
sprt @ 10+0.1 th 1 Try again this slower-but-maybe-more-accurate see() against new master.
16-10-06 Elb close_knights diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 47585 W: 8317 L: 8244 D: 31024
sprt @ 10+0.1 th 1 Take 3: bigger bonus
16-10-06 sni score_captures diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 9608 W: 1649 L: 1734 D: 6225
sprt @ 10+0.1 th 1 Take 2: only for PAWN
16-10-06 Voy evasionCMH diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29720 W: 5183 L: 5184 D: 19353
sprt @ 10+0.1 th 1 Try to use CMH stats for new simplified evasion scoring.
16-10-06 sni score_captures diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12768 W: 2217 L: 2288 D: 8263
sprt @ 10+0.1 th 1 In captures ordering, bonus for captures on the opponent king.
16-10-06 fau tuned diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 34727 W: 6172 L: 6211 D: 22344
sprt @ 10+0.1 th 1 Tuned patch
16-10-06 atu see diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 26040 W: 4675 L: 4442 D: 16923
sprt @ 10+0.1 th 1 Optimisation of Position::see and Position::see_sign
16-10-06 sto non_quiet_ttMove_bonus diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4021 W: 701 L: 810 D: 2510
sprt @ 10+0.1 th 1 Rationale for this is that non-quiet moves are usually more valuable than quiet ones in move ordering
16-10-06 Voy lmrt diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 20089 W: 3542 L: 3583 D: 12964
sprt @ 10+0.1 th 1 stc
16-10-06 Voy evasionSimple diff
LLR: -1.70 (-2.94,2.94) [-3.00,1.00]
Total: 13138 W: 2274 L: 2386 D: 8478
sprt @ 10+0.1 th 1 stc
16-10-06 pb0 correct_dc diff
ELO: -2.43 +-3.3 (95%) LOS: 7.6%
Total: 15000 W: 2636 L: 2741 D: 9623
15000 @ 10+0.1 th 1 Take 2: as queue is rather empty now, I like to see if take1 was just a lucky one or if effectively helps in handling discovered attacks on the queen correctly.
16-10-06 luc pawnattacksspan_initiat diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 87586 W: 15479 L: 15783 D: 56324
sprt @ 10+0.1 th 1 Last try (replacing OR with XOR), this time tested as a simplification (at least at STC)
16-10-06 Roc KnightCanJump diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9555 W: 1606 L: 1691 D: 6258
sprt @ 10+0.1 th 1 Smaller penalty
16-10-06 aji probcut_cond3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12823 W: 2265 L: 2336 D: 8222
sprt @ 10+0.1 th 1 Take 3: STC
16-10-05 Roc KnightCanJump diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22234 W: 3952 L: 3983 D: 14299
sprt @ 10+0.1 th 1 Bonus for knights which can safely attack a weak pawn on the next move.
16-10-05 sni qsearch_levels diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 6851 W: 1165 L: 1298 D: 4388
sprt @ 10+0.1 th 1 Use DEPTH_QS_NO_CHECKS = -2. Tested against passed "score_evasion" patch. This sort of makes sense: we can afford more checks in qsearch() since evasion scoring is now faster.
16-10-05 luc pawnattacksspan_initiat diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 18791 W: 3268 L: 3361 D: 12162
sprt @ 10+0.1 th 1 Take 2 (yielding values closer to the current ones)
16-10-05 aji probcut_condition_fix diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 50741 W: 9012 L: 8924 D: 32805
sprt @ 10+0.1 th 1 Fix a bug in previous test: STC
16-10-05 aji probcut_cond2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 31526 W: 5580 L: 5573 D: 20373
sprt @ 10+0.1 th 1 Take 2: STC
16-10-05 luc can_chain diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 37875 W: 6764 L: 6729 D: 24382
sprt @ 10+0.1 th 1 Four types of backward pawns. Penalties tuned with SPSA.
16-10-05 Elb close_knights diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22733 W: 3980 L: 4010 D: 14743
sprt @ 10+0.1 th 1 Small bonus for knights on adjacent squares, take 2
16-10-05 luc pawnattacksspan_initiat diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 21006 W: 3717 L: 3802 D: 13487
sprt @ 10+0.1 th 1 Initiative based on pawn attacks span, take 1. (in fact a simplification, but I prefer to avoid testing for no regression)
16-10-05 pb0 correct_dc diff
ELO: 1.70 +-4.1 (95%) LOS: 79.5%
Total: 10000 W: 1800 L: 1751 D: 6449
10000 @ 10+0.1 th 1 It is a little bit astonishing that as discovered check candidates we also consider pieces which do check oneself. How many ELO we loose by fixing this?
16-10-05 sni score_evasions diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 29333 W: 3809 L: 3700 D: 21824
sprt @ 60+0.6 th 1 LTC: variant 1 of atumanian patch for evasion scorin (simplify away SEE in evasion scoring)
16-10-05 SC stonewallSN diff
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 12715 W: 2322 L: 2144 D: 8249
sprt @ 10+0.1 th 1 Try tuned stonewall boni.
16-10-05 luc doubled_opposed diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12023 W: 2105 L: 2179 D: 7739
sprt @ 10+0.1 th 1 Retry applying different penalties for doubled pawns based on the opposed flag (I believe this was last attempted with scores by file...)
16-10-04 SC stonewallSN_tune diff
40722/50000 iterations
84314/100000 games played
100000 @ 10+0.1 th 1 Respin stonewall tuning on top of passed stonewall3 patch and including suggestion by Rocky
16-10-05 Voy fptt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 23410 W: 4148 L: 4174 D: 15088
sprt @ 10+0.1 th 1 take 2
16-10-05 jos pawn_tuning1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 4253 W: 741 L: 884 D: 2628
sprt @ 10+0.1 th 1 Another tuning try.
16-10-05 Elb close_knights diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 27120 W: 4866 L: 4876 D: 17378
sprt @ 10+0.1 th 1 Small bonus for knights on adjacent squares
16-10-05 jos pawn_push_ext diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13013 W: 2309 L: 2379 D: 8325
sprt @ 10+0.1 th 1 Pawn push extension.
16-10-04 luc can_chain_tune diff
29528/30000 iterations
60000/60000 games played
60000 @ 10+0.1 th 1 Since my latest patch looked good, try tuning those values, including one more backward pawn type...
16-10-05 aji probcut_condition diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17800 W: 3101 L: 3151 D: 11548
sprt @ 10+0.1 th 1 Don't do probcut if a previous search did not return more than rbeta: STC
16-10-05 aji probcut_depth diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 17220 W: 3029 L: 3127 D: 11064
sprt @ 10+0.1 th 1 Since low depth probcut failed quickly, test if we are using the optimal depth for probcut: STC
16-10-04 jos pruning_tweak diff
ELO: -3.60 +-2.9 (95%) LOS: 0.8%
Total: 20000 W: 3555 L: 3762 D: 12683
20000 @ 10+0.1 th 1 Measure the impact of no pruning if we have no pawns or if we have only pawns. This clearly helps with issue #760 and at mate-finding. (Idea from Texel.)
16-10-04 fau levers diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 46867 W: 8306 L: 8305 D: 30256
sprt @ 10+0.1 th 1 Levers additional rank
16-10-04 Voy fptt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 45737 W: 8197 L: 8129 D: 29411
sprt @ 10+0.1 th 1 Use a smaller futile margin if there is no ttMove.
16-10-03 jos safe_checks diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 83641 W: 14618 L: 14914 D: 54109
sprt @ 10+0.1 th 1 My try to simplify SafeCheck. This is kind of difficult because this highly depends on the already computed KingDanger value.
16-10-04 sni score_evasions diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 43889 W: 7849 L: 7767 D: 28273
sprt @ 10+0.1 th 1 Variant 1 of atumanian patch for evasion scoring: simplify away SEE in scoring
16-10-04 aji fpeval diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 49799 W: 8818 L: 8807 D: 32174
sprt @ 10+0.1 th 1 Return eval as it is instead of deducting futlity margin: STC Testing as sprt(0,4) even though it removes some logic.
16-10-04 Voy iids diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 50024 W: 8746 L: 8986 D: 32292
sprt @ 10+0.1 th 1 iid simplification (fixed bench)
16-10-04 sni score_evasions diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12942 W: 2266 L: 2336 D: 8340
sprt @ 10+0.1 th 1 Variant 2 of atumanian patch for evasion scoring
16-10-04 Elb minor_behind diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 21300 W: 3759 L: 3794 D: 13747
sprt @ 10+0.1 th 1 Bonus for close proximity of knight to enemy king
16-10-04 Voy inCheckRazoring diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 27907 W: 5027 L: 5034 D: 17846
sprt @ 10+0.1 th 1 Correct logic