Stockfish Testing Queue

Finished - 20009 tests

05-09-13 gl movecount_potential diff
LLR: -2.95 (-2.94,2.94)
Total: 2470 W: 449 L: 549 D: 1472
sprt @ 15+0.05 th 1 Try limiting movecount pruning by move potential score
05-09-13 jk qsbeta diff
LLR: -2.95 (-2.94,2.94)
Total: 11890 W: 2424 L: 2498 D: 6968
sprt @ 15+0.05 th 1 Small tweak to qsearch bestValue
05-09-13 jk qs0 diff
LLR: 2.97 (-2.94,2.94)
Total: 17468 W: 3698 L: 3540 D: 10230
sprt @ 15+0.05 th 1 Qsearch checks only at depth = 0.
05-09-13 gl threat_move diff
LLR: -2.95 (-2.94,2.94)
Total: 7708 W: 1556 L: 1642 D: 4510
sprt @ 15+0.05 th 1 No depth limit on threat move detection
05-09-13 sg lmr_conditions diff
LLR: -2.95 (-2.94,2.94)
Total: 1773 W: 321 L: 424 D: 1028
sprt @ 15+0.05 th 1 LMR: reduce non-promotion captures if see < 0. (Thanks to lucas!)
05-09-13 gl no_progress diff
LLR: -2.95 (-2.94,2.94)
Total: 28451 W: 5701 L: 5729 D: 17021
sprt @ 15+0.05 th 1 Idea from critter - prune rep moves when ahead (fixed bench issue take 2)
05-09-13 tv Storm diff
LLR: -2.96 (-2.94,2.94)
Total: 6967 W: 1207 L: 1276 D: 4484
sprt @ 60+0.05 th 1 LTC of combined storm rank 3 tune down
05-09-13 ur reduce_lmr diff
LLR: -2.96 (-2.94,2.94)
Total: 24127 W: 4728 L: 4769 D: 14630
sprt @ 15+0.05 th 1 this time I only not reduce the second check in a row and (ss-1)->staticEval!=VALUE_NONE means no checks 2 plies earlier(it was one ply earlier before we made the move and making the move without search does not change ss to ss+1)
05-09-13 lb allows diff
LLR: -2.96 (-2.94,2.94)
Total: 27311 W: 4776 L: 4744 D: 17791
sprt @ 60+0.05 th 1 Tweak allows()
05-09-13 lb cancel_reduction diff
LLR: -2.96 (-2.94,2.94)
Total: 10180 W: 2027 L: 2106 D: 6047
sprt @ 15+0.05 th 1 cancel reduction explicitely, rather than using a fail low hack. should save some unnecessary preliminary work in the research at full depth.
05-09-13 lb lmp diff
LLR: -2.95 (-2.94,2.94)
Total: 14653 W: 3036 L: 3102 D: 8515
sprt @ 15+0.05 th 1 LMP when depth < 12 plies only
02-09-13 mc skill_level^^ diff
ELO: 89.36 +-4.7 (95%) LOS: 100.0%
Total: 19387 W: 10841 L: 5962 D: 2584
20000 @ 30+0.05 th 1 Reschedule Skill Level 8 vs 7 with longer TC
05-09-13 ur reduce_lmr diff
LLR: -2.95 (-2.94,2.94)
Total: 8448 W: 1730 L: 1814 D: 4904
sprt @ 15+0.05 th 1 not reduce checks in LMR
05-09-13 tv Storm diff
LLR: -2.96 (-2.94,2.94)
Total: 39463 W: 8072 L: 8069 D: 23322
sprt @ 15+0.05 th 1 And more...
02-09-13 mc skill_level^ diff
ELO: 92.64 +-4.5 (95%) LOS: 100.0%
Total: 20000 W: 11061 L: 5851 D: 3088
20000 @ 30+0.05 th 1 Reschedule Skill Level 9 vs 8 with longer TC
05-09-13 tv Storm diff
LLR: 2.97 (-2.94,2.94)
Total: 8736 W: 1906 L: 1772 D: 5058
sprt @ 15+0.05 th 1 More tuning of rank 3
05-09-13 tv Storm diff
LLR: 2.97 (-2.94,2.94)
Total: 5132 W: 1126 L: 1003 D: 3003
sprt @ 15+0.05 th 1 Tune storm
03-09-13 ra allows diff
LLR: 2.97 (-2.94,2.94)
Total: 80260 W: 16445 L: 16113 D: 47702
sprt @ 15+0.05 th 1 Tweak allows()
05-09-13 lb qs_check diff
LLR: 2.96 (-2.94,2.94)
Total: 22972 W: 4124 L: 4020 D: 14828
sprt @ 60+0.05 th 1 qsearch simplification: do not prune useless checks (passed short TC)
04-09-13 lb qs_check diff
LLR: 2.96 (-2.94,2.94)
Total: 6243 W: 1302 L: 1195 D: 3746
sprt @ 15+0.05 th 1 qsearch simplification: do not prune useless checks
04-09-13 jh nonzero_outposts diff
LLR: -2.95 (-2.94,2.94)
Total: 14007 W: 2854 L: 2922 D: 8231
sprt @ 15+0.05 th 1 Gain outpost bonuses on all squares rank 3-6.
04-09-13 mc MIBishop_pair diff
LLR: -2.97 (-2.94,2.94)
Total: 23109 W: 4104 L: 4092 D: 14913
sprt @ 60+0.05 th 1 LTC: Real bishop pair() use in material imbalance. Reduces bench by 4 nodes but was an increase for 20130830 master. Minor fix, maybe there are more cases in material.cpp where this should be done?
04-09-13 gl null_threat diff
LLR: -2.95 (-2.94,2.94)
Total: 8776 W: 1825 L: 1908 D: 5043
sprt @ 15+0.05 th 1 Another take on Lucas null threat reduction nullification idea
04-09-13 jk wb diff
LLR: -2.96 (-2.94,2.94)
Total: 8495 W: 1693 L: 1777 D: 5025
sprt @ 15+0.05 th 1 Bonus for having dark squared bishop
04-09-13 jk wb diff
LLR: -2.96 (-2.94,2.94)
Total: 15544 W: 3167 L: 3231 D: 9146
sprt @ 15+0.05 th 1 Penalty for Darksquared bishop
04-09-13 lb 23e4eba13faac5ae75c3821 diff
LLR: -2.95 (-2.94,2.94)
Total: 12938 W: 2663 L: 2771 D: 7504
sprt @ 15+0.05 th 1 outpost simplification: value = 30
04-09-13 lb 4fcd0fe612f19da76acda5b diff
LLR: -2.95 (-2.94,2.94)
Total: 5665 W: 1129 L: 1237 D: 3299
sprt @ 15+0.05 th 1 simplified outpost: value = 10
04-09-13 lb e7be9209fd68440ffb8cfbd diff
LLR: -2.95 (-2.94,2.94)
Total: 27652 W: 5720 L: 5828 D: 16104
sprt @ 15+0.05 th 1 simplified outpost: value = 20
04-09-13 lb 436c9f6abadbf4f6fde87c6 diff
LLR: -2.96 (-2.94,2.94)
Total: 18964 W: 3285 L: 3389 D: 12290
sprt @ 60+0.05 th 1 null threat simplification: take 3 (passed short TC)
04-09-13 gl qs5 diff
LLR: -2.97 (-2.94,2.94)
Total: 10940 W: 1866 L: 1916 D: 7158
sprt @ 60+0.05 th 1 Joona LTC - QS checks: queen contact check
03-09-13 jk no_tt_qsearch diff
ELO: -52.46 +-3.2 (95%) LOS: 0.0%
Total: 19312 W: 2708 L: 5602 D: 11002
20000 @ 15+0.05 th 1 Measure value of TT access in qsearch
04-09-13 jh space_mask diff
LLR: -2.99 (-2.94,2.94)
Total: 30519 W: 6259 L: 6282 D: 17978
sprt @ 15+0.05 th 1 Extend space mask to B and G files.
03-09-13 ur addtime diff
LLR: -2.95 (-2.94,2.94)
Total: 2793 W: 539 L: 639 D: 1615
sprt @ 15+0.05 th 1 this time I decided to test the same idea with reducing slowmover(even if the previous moretime pass the second moretime may pass faster and I prefer to test more than one option before testing at longer time control)
03-09-13 ur addtime diff
LLR: -2.95 (-2.94,2.94)
Total: 1999 W: 367 L: 469 D: 1163
sprt @ 15+0.05 th 1 add more time for the instable pv when there is a faillow(maybe I need to compensate for it but I try first only one change and if it does not work I will try to use smaller number for slowmover as compensation
03-09-13 ra candidate diff
LLR: -2.96 (-2.94,2.94)
Total: 6351 W: 1260 L: 1350 D: 3741
sprt @ 15+0.05 th 1 Don't count candidate pawn supported by passed pawn. Far from perfect.
03-09-13 mc tweaks diff
LLR: -2.97 (-2.94,2.94)
Total: 31068 W: 5369 L: 5320 D: 20379
sprt @ 60+0.05 th 1 LTC: LMR >= 3 ply
03-09-13 lb contempt diff
LLR: -0.98 (-2.94,2.94)
Total: 14974 W: 3211 L: 3247 D: 8516
sprt @ 15+0.05 th 1 orthodox contempt, default 15cp
03-09-13 lb psq diff
LLR: -2.96 (-2.94,2.94)
Total: 3237 W: 592 L: 699 D: 1946
sprt @ 15+0.05 th 1 PST simplification: move constants to piece values instead of PST
03-09-13 lb 436c9f6abadbf4f6fde87c6 diff
LLR: 2.95 (-2.94,2.94)
Total: 15387 W: 3256 L: 3148 D: 8983
sprt @ 15+0.05 th 1 null threat simplification: take 3
03-09-13 lb 9e562787cb46e623b8666b7 diff
LLR: -2.96 (-2.94,2.94)
Total: 3538 W: 678 L: 786 D: 2074
sprt @ 15+0.05 th 1 null threat simplification: take 2
03-09-13 lb 8cb988b45cde01663937549 diff
LLR: -2.96 (-2.94,2.94)
Total: 36345 W: 7446 L: 7554 D: 21345
sprt @ 15+0.05 th 1 null threat simplification
03-09-13 sg lmr_conditions diff
LLR: -2.96 (-2.94,2.94)
Total: 7047 W: 1398 L: 1486 D: 4163
sprt @ 15+0.05 th 1 Correction LMR: do not reduce if move refutes threat move. The call of refutes have to be occur before do_move otherwise this is meaningless!
03-09-13 jk qs6 diff
LLR: -2.98 (-2.94,2.94)
Total: 2429 W: 455 L: 557 D: 1417
sprt @ 15+0.05 th 1 qs checks: double threats
03-09-13 jk qs5 diff
LLR: 2.95 (-2.94,2.94)
Total: 9927 W: 2128 L: 1992 D: 5807
sprt @ 15+0.05 th 1 QS checks: queen contact check
03-09-13 jk qs4 diff
LLR: -2.95 (-2.94,2.94)
Total: 2837 W: 522 L: 621 D: 1694
sprt @ 15+0.05 th 1 qs checks: one escape square
03-09-13 jk qs3 diff
LLR: -2.95 (-2.94,2.94)
Total: 15479 W: 3139 L: 3203 D: 9137
sprt @ 15+0.05 th 1 qs checks: Remove static eval check
03-09-13 jk qs2 diff
LLR: -2.95 (-2.94,2.94)
Total: 13607 W: 2800 L: 2869 D: 7938
sprt @ 15+0.05 th 1 qs checks 0
02-09-13 mc pawn_tune diff
ELO: -0.30 +-2.2 (95%) LOS: 39.7%
Total: 38857 W: 8023 L: 8056 D: 22778
40000 @ 15+0.05 th 1 Retest at fixed games: 1.25x candidate passed
03-09-13 rt SwordFish diff
ELO: -18.80 +-3.3 (95%) LOS: 0.0%
Total: 19484 W: 3903 L: 4956 D: 10625
20000 @ 15+0.05 th 1 Test SwordFish against Master. Since the -34 ELO try, I have reverted a bad mobility tune, tuned the knight outposts and narrowed the focus of the project by restoring endgame PSQT (in simplified form). This time, the end ELO difference should be close to -15. NOTE: This will be the last test relating to SwordFish unless Marco gives me the go-ahead, but it is critical to know how far off this version truly is, especially for the ability to make informed decisions.
03-09-13 jk qs1 diff
LLR: -2.96 (-2.94,2.94)
Total: 8013 W: 1354 L: 1418 D: 5241
sprt @ 60+0.05 th 1 QS recaptures -3 plies (test directly at long TC)