Stockfish Testing Queue

Finished - 35494 tests

16-09-28 Voy trf diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 12720 W: 2224 L: 2337 D: 8159
sprt @ 10+0.1 th 1 Reduction Formula Tweak
16-09-28 fau rooktweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 14638 W: 2524 L: 2631 D: 9483
sprt @ 10+0.1 th 1 Rook tweak after 78k spsa games
16-09-28 sni both_flanks9 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11556 W: 2017 L: 2093 D: 7446
sprt @ 10+0.1 th 1 Both flanks bonus = S(0, 10)
16-09-28 jos fmct diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 17446 W: 3036 L: 3133 D: 11277
sprt @ 10+0.1 th 1 FutilityMoveCounts tweak.
16-09-28 sni both_flanks9 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7184 W: 1269 L: 1364 D: 4551
sprt @ 10+0.1 th 1 Both flanks bonus = S(0, 30)
16-09-28 Voy statBonus diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7608 W: 1340 L: 1433 D: 4835
sprt @ 10+0.1 th 1 stc
16-09-28 pb0 historyFormula4' diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 36110 W: 4582 L: 4585 D: 26943
sprt @ 60+0.6 th 1 LTC: Take 4, linear coefficient = 10
16-09-28 Voy statPenaltyD2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3841 W: 636 L: 745 D: 2460
sprt @ 10+0.1 th 1 Take 2.
16-09-25 SC KRPPKRPagain diff
LLR: -1.69 (-2.94,2.94) [-3.00,1.00]
Total: 66560 W: 8472 L: 8642 D: 49446
sprt @ 60+0.6 th 1 New attempt at reducing degrees of freedom of KRPPKRP endgame specialized eval. Bench does not change but I have verified that it changes for the specific endgame. LTC. Before approving please read github discussion.
16-09-28 pb0 historyFormula4' diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 15337 W: 2735 L: 2548 D: 10054
sprt @ 10+0.1 th 1 Take 4, trying linear coefficient = 10 before going LTC with take 3
16-09-28 pb0 historyFormula4 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 17240 W: 3122 L: 2926 D: 11192
sprt @ 10+0.1 th 1 Take 3 (trying linear coefficient = 8)
16-09-27 luc same_color_supported diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 95700 W: 17357 L: 17076 D: 61267
sprt @ 10+0.1 th 1 (fixed base signature) Tuned values (new implementation: 1.2% faster here...)
16-09-28 pb0 historyFormula4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 39066 W: 6931 L: 6892 D: 25243
sprt @ 10+0.1 th 1 Take 2: Increase linear coefficient in history bonus formula (and compensate in exponent to maintain the same ceiling)
16-09-27 luc simple_psqt diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 59144 W: 10516 L: 10473 D: 38155
sprt @ 10+0.1 th 1 I finally also tuned queen's PSQT, is this now an improvement? I hope not a regression anyway!
16-09-28 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25598 W: 4565 L: 4582 D: 16451
sprt @ 10+0.1 th 1 less penalty. Take 4
16-09-27 sni supported_levers4 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 50367 W: 8902 L: 8816 D: 32649
sprt @ 10+0.1 th 1 Retry supported levers
16-09-27 SC stonewall2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 50151 W: 8907 L: 8822 D: 32422
sprt @ 10+0.1 th 1 Try to revive the old center bind in the form of a stonewall.
16-09-25 Elb outpost_rank diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 68190 W: 12214 L: 12052 D: 43924
sprt @ 10+0.1 th 1 Rank dependent outpost value
16-09-27 Elb outpost_rank2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 34884 W: 6149 L: 6128 D: 22607
sprt @ 10+0.1 th 1 Rank dependent outpost value. Contains values from latest tuning attempt.
16-09-27 Voy statPenaltyD diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 39609 W: 7019 L: 6978 D: 25612
sprt @ 10+0.1 th 1 Extra Penalty if the prior TTmove is also a PvNode.
16-09-27 sg king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16007 W: 2835 L: 2892 D: 10280
sprt @ 10+0.1 th 1 Penalty +50%. Take 3
16-09-27 sni discov_checks_in_see6 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11903 W: 2103 L: 2178 D: 7622
sprt @ 10+0.1 th 1 Simpler try
16-09-27 SC rookEndgames diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14510 W: 2499 L: 2563 D: 9448
sprt @ 10+0.1 th 1 Both tests with scale factor reduction only in absence of passers performed worse than the corresponding ones without the tweak. Are we overestimating passers in KRPsKRPs?
16-09-27 jos no_lmp_after_null diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5460 W: 914 L: 1016 D: 3530
sprt @ 10+0.1 th 1 Don't allow movecount-pruning after a null-move. Most probably this is no problem, but let's see.
16-09-27 Elb tune_outpostrank diff
28140/30000 iterations
57903/60000 games played
60000 @ 20+0.2 th 1 Redo the tuning with same ck value for all parameters (thanks lucabrivio)
16-09-27 mbo simple_storm_eval diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 14007 W: 2689 L: 2875 D: 8443
sprt @ 10+0.1 th 1 Simple storm eval.
16-09-27 sg king_safety diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 36130 W: 6466 L: 6438 D: 23226
sprt @ 10+0.1 th 1 Penalty if safe checks exists and the king has no square for retreat. Exclude queen contact checks, that's possible too much. Take 2
16-09-26 luc simple_psqt_tune_Q diff
19482/20000 iterations
40000/40000 games played
40000 @ 20+0.2 th 1 Last piece to tune PSQT for is the queen, I hope the positive trend won't stop!
16-09-27 Voy statPenaltyD diff
LLR: -0.46 (-2.94,2.94) [0.00,5.00]
Total: 163 W: 24 L: 43 D: 96
sprt @ 10+0.1 th 1 Apply stat penalty if the prior move is a pvNode.
16-09-27 sg king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17736 W: 3148 L: 3198 D: 11390
sprt @ 10+0.1 th 1 Penalty if safe checks exists and the king has no square for retreat
16-09-27 pb0 see_promotion_fix diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11746 W: 2048 L: 2124 D: 7574
sprt @ 10+0.1 th 1 Promotion in SEE: test whether this fix helps or hurts SF's playing strength
16-09-27 luc same_color_supported diff
LLR: -1.37 (-2.94,2.94) [0.00,5.00]
Total: 8041 W: 1433 L: 1457 D: 5151
sprt @ 10+0.1 th 1 Tuned values
16-09-27 Elb outpost_rank2 diff
LLR: -1.47 (-2.94,2.94) [0.00,5.00]
Total: 5195 W: 925 L: 965 D: 3305
sprt @ 10+0.1 th 1 Rank dependent outpost value. Take 2 with tuned values.
16-09-26 luc same_color_supported_tu diff
28977/30000 iterations
60000/60000 games played
60000 @ 10+0.1 th 1 Looks like the burden on performance outweighs any possible benefit of my patches, but how could those values possibly be set? Really speculative, so low throughput...
16-09-26 Elb tune_outpostrank diff
29075/30000 iterations
59745/60000 games played
60000 @ 20+0.2 th 1 As test http://tests.stockfishchess.org/tests/view/57e7d8750ebc59763f358e59 looks promising, but struggles, lower the priority of that test and tune the values for the outposts.
16-09-26 Voy cftft diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7036 W: 1191 L: 1286 D: 4559
sprt @ 10+0.1 th 1 Try color-from-to-from-to...being memory conscious by dividing board up by 16 squares. Take 1
16-09-26 Voy cft_fm diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4712 W: 783 L: 888 D: 3041
sprt @ 10+0.1 th 1 Try color-from-to for followup moves.
16-09-26 sni protected_outpost' diff
LLR: -3.06 (-2.94,2.94) [0.00,4.00]
Total: 43979 W: 5718 L: 5756 D: 32505
sprt @ 60+0.6 th 1 LTC: Protected reachable outposts
16-09-26 jos singular diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 42551 W: 7520 L: 7533 D: 27498
sprt @ 10+0.1 th 1 Don't let rbeta exceed beta.
16-09-24 fau tune diff
38906/40000 iterations
80000/80000 games played
80000 @ 20+0.2 th 1 Continuing with the tuning of the mgvalues (Fixed hash)
16-09-26 SC rookEndgames diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16397 W: 2776 L: 2833 D: 10788
sprt @ 10+0.1 th 1 Same as last attempt, but only if strong side has no passed pawns.
16-09-26 SC rookEndgames diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18502 W: 3228 L: 3275 D: 11999
sprt @ 10+0.1 th 1 Reduce by 10% and then call it a day until new amazing ideas arise :-)
16-09-26 sni protected_outpost' diff
LLR: -1.74 (-2.94,2.94) [0.00,5.00]
Total: 15866 W: 2794 L: 2801 D: 10271
sprt @ 10+0.1 th 1 Protected reachable outposts: try a version with popcount
16-09-26 Voy improving2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 7941 W: 1355 L: 1447 D: 5139
sprt @ 10+0.1 th 1 if depth = 1 improving = true
16-09-26 Roc WeakRook diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10362 W: 1809 L: 1890 D: 6663
sprt @ 10+0.1 th 1 Another try on weak rook
16-09-26 luc same_color_supported diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16635 W: 2952 L: 3007 D: 10676
sprt @ 10+0.1 th 1 Try considering opponent's pawns too (not sure how this should be!), low throughput
16-09-26 Voy improving diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9424 W: 1605 L: 1690 D: 6129
sprt @ 10+0.1 th 1 improving = false if eval is way below alpha at low depth.
16-09-26 pb0 discov_checks_in_see6 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11016 W: 1917 L: 1996 D: 7103
sprt @ 10+0.1 th 1 Take 6 (and probably last): bitboard approach to generally handle discoverings, variant.
16-09-25 luc same_color_supported diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 21179 W: 3847 L: 3882 D: 13450
sprt @ 10+0.1 th 1 Perhaps if pawns on same color square of bishops are either supported or supporting, that could be less advantageous later in the game...
16-09-25 luc simple_psqt diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 53945 W: 9590 L: 9565 D: 34790
sprt @ 10+0.1 th 1 ...now with tuned bishop PSQT!