Stockfish Testing Queue

Finished - 59227 tests

13-03-26 mco smooth_passed_pawn diff
ELO: -4.51 +-4.4 (95%) LOS: 2.2%
Total: 8087 W: 1318 L: 1423 D: 5346
16000 @ 60+0.05 th 1
Third attempt by Jundery. This time seems good at long TC. Let's verify it.
13-03-26 jki equal_see_pruning_depth diff
ELO: 4.19 +-3.5 (95%) LOS: 99.1%
Total: 15000 W: 3014 L: 2833 D: 9153
16000 @ 15+0.05 th 1
Prune moves with equal SEE also at depth zero.
13-03-27 lbr lucas diff
ELO: -1.41 +-3.4 (95%) LOS: 20.6%
Total: 16000 W: 3097 L: 3162 D: 9741
16000 @ 15+0.05 th 1
restest without regressive LMR smoothing at shallow depth part
13-03-28 lbr lucas diff
ELO: -3.80 +-3.5 (95%) LOS: 1.8%
Total: 14351 W: 2717 L: 2874 D: 8760
16000 @ 15+0.05 th 1
retest without the regressive new reduction formula
13-03-28 gli all_cut_iid diff
ELO: -0.00 +-5.0 (95%) LOS: 50.0%
Total: 7271 W: 1438 L: 1438 D: 4395
16000 @ 15+0.05 th 1
Try a depth limited version of (depth - 4) reduction
13-03-28 jki null_move3 diff
ELO: -3.69 +-3.5 (95%) LOS: 2.0%
Total: 14707 W: 2824 L: 2980 D: 8903
16000 @ 15+0.05 th 1
Null move: R = 4 * ONE_PLY + depth / 6 + ONE_PLY(?)
13-03-29 mco equal_see_pruning_depth diff
ELO: -1.43 +-3.4 (95%) LOS: 20.4%
Total: 13362 W: 2186 L: 2241 D: 8935
16000 @ 60+0.05 th 1
Rescheduled: Prune moves with equal SEE also at depth zero.
13-03-29 jki prune_capture_neg_see diff
ELO: -4.31 +-4.4 (95%) LOS: 2.8%
Total: 9441 W: 1820 L: 1937 D: 5684
16000 @ 15+0.05 th 1
Prune bad captures in search() when newdepth < ONE_PLY
13-03-29 gli see_search diff
ELO: -1.35 +-3.4 (95%) LOS: 21.9%
Total: 15718 W: 3073 L: 3134 D: 9511
16000 @ 15+0.05 th 1
Less aggressive LMR since we are now reducing bad see moves more.
13-03-29 mco increase_iid diff
ELO: 4.08 +-3.1 (95%) LOS: 99.5%
Total: 16000 W: 2742 L: 2554 D: 10704
16000 @ 60+0.05 th 1
Set IID half way between d/2 and d-4. Test directly at long TC because short is not sensible
13-03-29 gli OutpostChangeTake2 diff
ELO: -0.46 +-4.2 (95%) LOS: 41.5%
Total: 10689 W: 2125 L: 2139 D: 6425
16000 @ 15+0.05 th 1
RyanTaker outpost changes take 2
13-03-30 lbr lucas_qsearch_pruning diff
ELO: -6.88 +-4.3 (95%) LOS: 0.1%
Total: 10057 W: 1909 L: 2108 D: 6040
16000 @ 15+0.05 th 1
remove "useless checks" and a couple of PV conditions
13-03-30 mco qs_no_recaptures diff
ELO: -3.24 +-3.5 (95%) LOS: 3.3%
Total: 13096 W: 2158 L: 2280 D: 8658
16000 @ 60+0.05 th 1
Reschedule due to bogus bench signature of master branch
13-03-30 gli last_chance diff
ELO: -9.51 +-4.5 (95%) LOS: 0.0%
Total: 9243 W: 1716 L: 1969 D: 5558
16000 @ 15+0.05 th 1
An experiment on merging together marginal patches
13-03-30 jki null_move4 diff
ELO: 3.56 +-3.3 (95%) LOS: 98.3%
Total: 16000 W: 3103 L: 2939 D: 9958
16000 @ 15+0.05 th 1
Raise depth limit of null move verification search to 10 plies
13-03-30 jki null_move5 diff
ELO: -1.43 +-3.4 (95%) LOS: 20.3%
Total: 16000 W: 3121 L: 3187 D: 9692
16000 @ 15+0.05 th 1
R = 3 * ONE_PLY + depth / 3 + ONE_PLY(?)
13-03-30 jki asymm_see diff
ELO: 2.82 +-3.3 (95%) LOS: 95.1%
Total: 16000 W: 3148 L: 3018 D: 9834
16000 @ 15+0.05 th 1
Asymmetric SEE pruning rule trick in qsearch()
13-03-30 jki symm_kingsafety diff
ELO: -14.23 +-7.5 (95%) LOS: 0.0%
Total: 3469 W: 670 L: 812 D: 1987
24000 @ 15+0.05 th 1
Simplification attempt: Symmetric king safety
13-03-30 mco master diff
ELO: 18.76 +-2.8 (95%) LOS: 100.0%
Total: 20000 W: 3903 L: 2824 D: 13273
20000 @ 60+0.05 th 1
Regression test after the last IID tweak
13-03-30 gli prune_more_checks diff
ELO: -17.24 +-6.8 (95%) LOS: 0.0%
Total: 4054 W: 719 L: 920 D: 2415
16000 @ 15+0.05 th 1
Prune more checks in q-search
13-04-02 gli move_horizon diff
ELO: -14.59 +-7.6 (95%) LOS: 0.0%
Total: 3264 W: 592 L: 729 D: 1943
16000 @ 15+0.05 th 1
Don't look as far ahead for move_horizon
13-03-31 lbr lucas_qsearch_pruning diff
ELO: 0.67 +-3.3 (95%) LOS: 65.3%
Total: 16000 W: 3110 L: 3079 D: 9811
16000 @ 15+0.05 th 1
remove some PV conditions in QS
13-03-31 jki double_gain diff
ELO: 3.45 +-3.4 (95%) LOS: 97.8%
Total: 16000 W: 3226 L: 3067 D: 9707
16000 @ 15+0.05 th 1
Double Impact of Gain tables. Very unorthodox idea.
13-03-31 jki prune_under_promo diff
ELO: 0.17 +-3.3 (95%) LOS: 54.1%
Total: 16000 W: 3065 L: 3057 D: 9878
16000 @ 15+0.05 th 1
Prune under promotions (excluding knight checks) when depth < 10 plies
13-03-31 jki orthodox_tt_depth diff
ELO: 5.28 +-2.8 (95%) LOS: 100.0%
Total: 24000 W: 4911 L: 4546 D: 14543
24000 @ 15+0.05 th 1
Set ttDepth = depth Accurary vs. speed
13-03-31 jki red_2plies diff
ELO: -38.15 +-3.4 (95%) LOS: 0.0%
Total: 16000 W: 2373 L: 4123 D: 9504
16000 @ 15+0.05 th 1
Try 2 * ONE_PLY reductions everywhere. Please do not stop the test even if it goes badly. I want to see how much the current complex implementation of LMR is worth.
13-03-31 mco null_move4 diff
ELO: 2.08 +-3.1 (95%) LOS: 90.9%
Total: 16000 W: 2641 L: 2545 D: 10814
16000 @ 60+0.05 th 1
Raise depth limit of null move verification search to 10 plies. retest at long TC
13-04-01 gli asymm_see diff
ELO: 1.39 +-3.1 (95%) LOS: 81.0%
Total: 16000 W: 2689 L: 2625 D: 10686
16000 @ 60+0.05 th 1
Long TC: Asymmetric SEE pruning rule trick in qsearch()
13-04-01 gli prune_more_checks diff
ELO: -3.48 +-3.4 (95%) LOS: 2.2%
Total: 15789 W: 3017 L: 3175 D: 9597
16000 @ 15+0.05 th 1
Try removing eval condition for pruning checks
13-04-01 gli old_dangerous_check diff
ELO: -3.37 +-3.4 (95%) LOS: 2.6%
Total: 16000 W: 3102 L: 3257 D: 9641
16000 @ 15+0.05 th 1
Try bringing back old pruning from dangerous_check
13-04-01 mco double_gain diff
ELO: 4.02 +-3.1 (95%) LOS: 99.5%
Total: 16000 W: 2729 L: 2544 D: 10727
16000 @ 60+0.05 th 1
Double Impact of Gain tables. Very unorthodox idea. Promote this guy to the long TC major league :-)
13-04-02 gli orthodox_tt_depth diff
ELO: -4.59 +-3.1 (95%) LOS: 0.2%
Total: 15681 W: 2446 L: 2653 D: 10582
16000 @ 60+0.05 th 1
LongTC. Set ttDepth = depth (Accurary vs. speed)
13-04-02 lbr lucas_qsearch_pruning diff
ELO: 0.87 +-3.4 (95%) LOS: 69.1%
Total: 16000 W: 3247 L: 3207 D: 9546
16000 @ 15+0.05 th 1
remove some more PV conditions
13-04-03 lbr lucas_verification_sear diff
ELO: -1.23 +-3.2 (95%) LOS: 22.8%
Total: 14700 W: 2417 L: 2469 D: 9814
16000 @ 60+0.05 th 1
remove verification search completely. test directly at long TC.
13-04-03 gli storm diff
ELO: 2.53 +-3.5 (95%) LOS: 91.9%
Total: 14709 W: 2977 L: 2870 D: 8862
16000 @ 15+0.05 th 1
Higher penalty for rank 6 opposed storm pawn (it restricts king mobility)
13-04-03 mco master diff
ELO: 15.44 +-2.8 (95%) LOS: 100.0%
Total: 20000 W: 3928 L: 3040 D: 13032
20000 @ 60+0.05 th 1
Regression test after Joona's "Double Gains"
13-04-03 gli master diff
ELO: 33.96 +-10.5 (95%) LOS: 100.0%
Total: 1714 W: 433 L: 266 D: 1015
sprt @ 5+0.05 th 1
SPRT test!
13-04-03 gli see_search diff
ELO: -30.32 +-11.9 (95%) LOS: 0.0%
Total: 1413 W: 244 L: 367 D: 802
sprt @ 5+0.05 th 1
SPRT: verification on a bad patch
13-04-03 gli master diff
ELO: -2.19 +-4.3 (95%) LOS: 16.0%
Total: 11925 W: 2802 L: 2877 D: 6246
sprt @ 2+0.05 th 1
SPRT: test against same version
13-04-04 lbr lucas_qsearch_pruning diff
ELO: 0.32 +-5.0 (95%) LOS: 55.1%
Total: 7546 W: 1523 L: 1516 D: 4507
sprt @ 15+0.05 th 1
remove more PV conditions
13-04-04 mco null_move4 diff
ELO: 2.41 +-2.0 (95%) LOS: 99.0%
Total: 43729 W: 8726 L: 8423 D: 26580
sprt @ 15+0.05 th 1
Raise depth limit of null move verification search to 10 plies. Retest using SPRT on this real case of interesting but still undecided patch.
13-04-04 gli storm diff
ELO: -1.99 +-3.6 (95%) LOS: 13.9%
Total: 12393 W: 2107 L: 2178 D: 8108
sprt @ 60+0.05 th 1
SPRT Long TC: Higher penalty for rank 6 opposed storm pawn (it restricts king mobility)
13-04-04 mco null_verification_8 diff
ELO: -0.09 +-2.6 (95%) LOS: 47.2%
Total: 25744 W: 5019 L: 5026 D: 15699
sprt @ 15+0.05 th 1
Experimental SPRT test: verify use of SPRT for fine tuning. In particular test if null_verification_8 is better than null_verification_10
13-04-04 mco null_verification_12 diff
ELO: 2.56 +-2.0 (95%) LOS: 99.5%
Total: 46568 W: 9240 L: 8897 D: 28431
sprt @ 15+0.05 th 1
Experimental SPRT test: verify use of SPRT for fine tuning. In particular test if null_verification_12 is better than null_verification_10
13-04-05 lbr lucas_qsearch_pruning diff
ELO: 8.31 +-6.0 (95%) LOS: 99.7%
Total: 5771 W: 1373 L: 1235 D: 3163
sprt @ 5+0.05 th 1
remove more PV conditions
13-04-05 lbr lucas_qsearch_pruning diff
ELO: 12.52 +-7.1 (95%) LOS: 100.0%
Total: 3470 W: 716 L: 591 D: 2163
sprt @ 15+0.05 th 1
remove more PV conditions
13-04-05 gli orthodox_tt_depth diff
ELO: -9.55 +-6.8 (95%) LOS: 0.3%
Total: 3929 W: 725 L: 833 D: 2371
sprt @ 15+0.05 th 1
SPRT retest, 4MB hash, Set ttDepth = depth
13-04-05 gli King_Saftey_A1 diff
ELO: 5.41 +-4.7 (95%) LOS: 98.8%
Total: 7965 W: 1588 L: 1464 D: 4913
sprt @ 15+0.05 th 1
RyanTakker idea - tweak pawn shelter based on presence of queen
13-04-05 mco jp_tweaks diff
ELO: 3.33 +-2.7 (95%) LOS: 99.2%
Total: 25866 W: 5461 L: 5213 D: 15192
sprt @ 15+0.05 th 1
UCI parameters tweak by JP: according to his tests this setup is stronger than default.
13-04-05 jki asymm_see2 diff
ELO: 3.53 +-2.8 (95%) LOS: 99.3%
Total: 18794 W: 3098 L: 2907 D: 12789
sprt @ 60+0.05 th 1
SPRT: Final judgement for asymm_see2