Stockfish Testing Queue

Finished - 22425 tests

19-12-14 sg king_block_pawn diff
ELO: -2.18 +-2.4 (95%) LOS: 3.7%
Total: 32659 W: 6502 L: 6707 D: 19450
40000 @ 15+0.05 th 1 big_king_safety: its unlikely but perhaps the change from 200 to 300 for king blocks pawn is responsible for the elo gain, so go for safety and test this version against the current master (see pull request comment)
19-12-14 jo KQKRPs_endgame diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 49388 W: 10093 L: 10024 D: 29271
sprt @ 15+0.05 th 1 Slightly changed downscaling to having more effect. I decided to only test take 2. See also commit notes, please.
19-12-14 jh krp_endgame diff
LLR: 4.44 (-2.94,2.94) [-3.00,1.00]
Total: 32847 W: 5554 L: 5375 D: 21918
sprt @ 60+0.05 th 1 KQKRPs endgame adjustment.
19-12-14 mb midgame_pp1 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14651 W: 2926 L: 2993 D: 8732
sprt @ 15+0.05 th 1 Raise midgame passed pawn values a bit.
18-12-14 pe tm diff
ELO: -0.66 +-2.5 (95%) LOS: 30.1%
Total: 30000 W: 5942 L: 5999 D: 18059
30000 @ 15+0.05 th 1 Decrease time available if for current iteration it looks like the best move would not change. Fix
19-12-14 jh krp_endgame diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 62936 W: 12648 L: 12604 D: 37684
sprt @ 15+0.05 th 1 KQKRPs endgame adjustment.
18-12-14 pe tm diff
ELO: -22.82 +-22.3 (95%) LOS: 2.2%
Total: 366 W: 60 L: 84 D: 222
30000 @ 15+0.05 th 1 Decrease time available if for current iteration it looks like the best move would not change
18-12-14 SC streamline_KPPKPP diff
23081/30000 iterations
41650/60000 games played
60000 @ 5+0.1 th 1 KPPKPP eval patch passed STC and failed LTC. Retry to tune a streamlined patch with a endgame-oriented tc.
17-12-14 lb tune diff
LLR: -4.38 (-2.94,2.94) [-3.00,1.00]
Total: 94218 W: 15734 L: 16115 D: 62369
sprt @ 60+0.05 th 1 LTC: simplified storm with locally tuned values
18-12-14 SC specKPPKPPeval diff
LLR: -3.73 (-2.94,2.94) [0.00,6.00]
Total: 11768 W: 1949 L: 2022 D: 7797
sprt @ 60+0.05 th 1 SPRT for a corrected version of KPP vs KPP eval patch. Previous version did not detect symmetric positions correctly.
17-12-14 sg big_king_safety diff
LLR: 2.96 (-2.94,2.94) [0.00,6.00]
Total: 10311 W: 1876 L: 1721 D: 6714
sprt @ 60+0.05 th 1 LTC: Measure big king safety tuning with corrected values
17-12-14 jh time_1 diff
LLR: -3.71 (-2.94,2.94) [-1.50,4.50]
Total: 47696 W: 9608 L: 9610 D: 28478
sprt @ 15+0.05 th 1 Reduce bonus time after consecutive fail highs.
17-12-14 sg big_king_safety diff
ELO: 3.46 +-2.2 (95%) LOS: 99.9%
Total: 40000 W: 8275 L: 7877 D: 23848
40000 @ 15+0.05 th 1 Measure big king safety tuning with corrected values (see forum). Prio -2
17-12-14 SC specKPPKPPeval diff
LLR: 3.46 (-2.94,2.94) [-1.50,4.50]
Total: 35549 W: 7328 L: 7103 D: 21118
sprt @ 15+0.05 th 1 SPRT for a corrected version of KPP vs KPP eval patch. Previous version did not detect symmetric positions correctly.
14-12-14 sg spsa_big_king_safety diff
47490/50000 iterations
90852/100000 games played
100000 @ 60+0.05 th 1 Big king safety tuning. Stormdanger and Shelterweakness indexed by file pairs (a/h,b/g,c/f,d/e). Special case where king blocks pawn is incorporated in Stormdanger. LTC because TC-dependant. There are 93 parameters so i use following SPSA configuration: Games=100000 Gamma=0.159 Alpha=0.558 C=5 (except in maxSafety C=10 is used) Prio -1
17-12-14 np DB-PVStability-2 diff
ELO: -0.33 +-5.7 (95%) LOS: 45.5%
Total: 5263 W: 958 L: 963 D: 3342
20000 @ 30+0.05 th 1 hypotheis: resolving pvstability at much higher depths won't help anymore. faster machines should be more sensitive to this change
16-12-14 np DB-PVStability diff
ELO: 0.55 +-3.0 (95%) LOS: 64.1%
Total: 18838 W: 3484 L: 3454 D: 11900
20000 @ 30+0.05 th 1 give more time for pv stability at higher depth.
16-12-14 lb tune diff
LLR: 3.26 (-2.94,2.94) [-3.50,0.50]
Total: 28113 W: 5754 L: 5655 D: 16704
sprt @ 15+0.05 th 1 simplified storm with locally tuned values
16-12-14 vi reduce_dithering2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 35383 W: 7183 L: 7192 D: 21008
sprt @ 15+0.05 th 1 Take 2 with a 2cp threshold as suggested by more than one commenter
16-12-14 Fi allowZeroKey diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 95565 W: 19098 L: 19114 D: 57353
sprt @ 15+0.05 th 1 Simplification. Don't treat 0 key as special.
16-12-14 SC specKPPKPPeval diff
ELO: 0.89 +-3.1 (95%) LOS: 71.5%
Total: 20000 W: 4075 L: 4024 D: 11901
20000 @ 15+0.05 th 1 In nonsymmetric KPP KPP endgames only reward advanced pawns. Showed some effect in superfast endgames. (And thank you to Arjun for the patience)
16-12-14 vi reduce_dithering2 diff
LLR: -3.81 (-2.94,2.94) [-1.50,4.50]
Total: 15189 W: 3079 L: 3175 D: 8935
sprt @ 15+0.05 th 1 Do not consider a new best move as indicating a critical position if the eval has only increased by <= 5cp. Suggested by consideration of engine output from TCEC example by Jose.
15-12-14 pe tm diff
ELO: 0.29 +-2.5 (95%) LOS: 59.1%
Total: 30000 W: 5938 L: 5913 D: 18149
30000 @ 15+0.05 th 1 Reduce available time if current iteration did not fail low at root. Take 3
16-12-14 pe tm diff
LLR: -3.30 (-2.94,2.94) [-3.00,1.00]
Total: 38248 W: 7459 L: 7708 D: 23081
sprt @ 15+0.05 th 1 Remove hard stop for unchanging root move. Test for no regression.
15-12-14 pe tm diff
LLR: -3.35 (-2.94,2.94) [0.00,6.00]
Total: 22022 W: 3648 L: 3659 D: 14715
sprt @ 60+0.05 th 1 LTC. Reduce available time if current iteration did not fail low at root.
15-12-14 pe tm diff
ELO: -0.06 +-2.5 (95%) LOS: 48.2%
Total: 30000 W: 5968 L: 5973 D: 18059
30000 @ 15+0.05 th 1 Remove hard stop for unchanging root move
15-12-14 pe tm diff
ELO: 0.42 +-2.5 (95%) LOS: 62.9%
Total: 30000 W: 5956 L: 5920 D: 18124
30000 @ 15+0.05 th 1 Reduce available time if current iteration did not fail low at root. Take 2
15-12-14 mc fix_tempo diff
LLR: 3.15 (-2.94,2.94) [-3.00,1.00]
Total: 49812 W: 10095 L: 10016 D: 29701
sprt @ 15+0.05 th 1 Fix: Don't account for Tempo in specialized endgames
15-12-14 sn LMR diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 38586 W: 7710 L: 7710 D: 23166
sprt @ 15+0.05 th 1 Take 3 (simpler version)
14-12-14 vi reduce_dithering diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 27065 W: 4539 L: 4512 D: 18014
sprt @ 60+0.05 th 1 Reschedule at LTC after STC sort pass. Acknowledged that if this works, there is a better way to express the change.
15-12-14 pe tm diff
LLR: 3.37 (-2.94,2.94) [-1.50,4.50]
Total: 28781 W: 5830 L: 5629 D: 17322
sprt @ 15+0.05 th 1 Reduce available time if current iteration did not fail low at root. Reschedule because of wrong options
15-12-14 Hj lower_maxratio diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 14543 W: 2916 L: 3023 D: 8604
sprt @ 15+0.05 th 1 Reduce max ratio to see if we can spend slightly less time on longer moves
14-12-14 gl bishop_blocked diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7820 W: 1615 L: 1701 D: 4504
sprt @ 15+0.05 th 1 Penalize bishops behind our own blocked pawns
14-12-14 np faster_initial diff
LLR: -3.05 (-2.94,2.94) [-1.50,4.50]
Total: 37088 W: 7559 L: 7566 D: 21963
sprt @ 15+0.05 th 1 test if first moves need less time.
14-12-14 gl LMR diff
LLR: -3.33 (-2.94,2.94) [-1.50,4.50]
Total: 4802 W: 889 L: 996 D: 2917
sprt @ 15+0.05 th 1 Take 2 (slightly less reduction for the improving branches) (Rescheduled by Gary after incorrect purging code got it)
14-12-14 pe tm diff
LLR: 1.55 (-2.94,2.94) [-1.50,4.50]
Total: 1930 W: 405 L: 344 D: 1181
sprt @ 15+0.05 th 1 Reduce available time if current iteration did not fail low at root
14-12-14 pe tm diff
LLR: -5.66 (-2.94,2.94) [-1.00,4.00]
Total: 45627 W: 9007 L: 9127 D: 27493
sprt @ 15+0.05 th 1 Reduce available time if did not fail low at root
14-12-14 sn LMR diff
LLR: -2.76 (-2.94,2.94) [-1.50,4.50]
Total: 2364 W: 441 L: 535 D: 1388
sprt @ 15+0.05 th 1 Take 2 (slightly less reduction for the improving branches)
13-12-14 sg history diff
LLR: -4.39 (-2.94,2.94) [-1.50,4.50]
Total: 19474 W: 3886 L: 3991 D: 11597
sprt @ 15+0.05 th 1 add half bonus if Max exceeded (so higher depth influence history)
14-12-14 sn LMR diff
LLR: -2.84 (-2.94,2.94) [-1.50,4.50]
Total: 4389 W: 850 L: 941 D: 2598
sprt @ 15+0.05 th 1 Tweak LMR reduction for non-improving branches
14-12-14 Fi ttSaveNoRead diff
ELO: -0.68 +-2.8 (95%) LOS: 31.7%
Total: 20000 W: 3347 L: 3386 D: 13267
20000 @ 60+0.05 th 1 LTC failed but was rescheduled w/ only 4MB. Since this patch affects cache behavior and there appears to be such a big difference between the STC and LTC result I would like to quickly measure LTC elo w/ 16MB while queue is empty anyway. Pri -4
14-12-14 vi reduce_dithering diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 10764 W: 2238 L: 2101 D: 6425
sprt @ 15+0.05 th 1 2nd (and probably last) try with a less aggressive decay factor. Rescheduled after a ticking-off from Arjun :)
14-12-14 Ma baseimportancetune diff
19716/20000 iterations
40000/40000 games played
40000 @ 15+0.05 th 1 Verify whether not having any ply-independent move importance is really optimal, and get some tuned value otherwise.
14-12-14 vi time_mgmt_enhance_try1 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 5869 W: 1127 L: 1218 D: 3524
sprt @ 15+0.05 th 1 Reworked and cleaned up time management patch. Based on fwittman's but simplified. Nabial Danial claimed Elo gain on 20,000 games - so trying while framework has some space.
14-12-14 Ro KingMobility_2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 5083 W: 988 L: 1081 D: 3014
sprt @ 15+0.05 th 1 Simplifcation on previous idea. Computing all safe squares in ring 2
14-12-14 Ro KingMobility diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 9057 W: 1762 L: 1844 D: 5451
sprt @ 15+0.05 th 1 Adding a rough measure of King 1 and two mobility in the King Safety evaluation.
13-12-14 jk ttSaveNoRead diff
LLR: -2.94 (-2.94,2.94) [0.00,6.00]
Total: 10029 W: 1663 L: 1717 D: 6649
sprt @ 60+0.05 th 1 Take 2: Reduce reading even further by making a local copy if we get a hit. (Rescheduled by Joona because queue was empty)
13-12-14 Fi insta_move diff
ELO: 1.64 +-2.5 (95%) LOS: 89.8%
Total: 30000 W: 6308 L: 6166 D: 17526
30000 @ 15+0.05 th 1 Rewritten to handle transpositions in the first two moves of the PV that still lead to the same stable 3rd move. Pri -2
13-12-14 Fi ttSaveNoRead diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11047 W: 2289 L: 2151 D: 6607
sprt @ 15+0.05 th 1 Take 2: Reduce reading even further by making a local copy if we get a hit.
13-12-14 vi reduce_dithering diff
LLR: -0.85 (-2.94,2.94) [-1.50,4.50]
Total: 24578 W: 5025 L: 4987 D: 14566
sprt @ 15+0.05 th 1 2nd (and probably last) try with a less aggressive decay factor.