Stockfish Testing Queue

Finished - 37193 tests

15-10-13 jhe simple_time diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 4771 W: 837 L: 1004 D: 2930
sprt @ 10+0.1 th 1 Simplify time management.
15-10-13 Fis TT_threadPriority diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5822 W: 864 L: 964 D: 3994
sprt @ 15+0.05 th 7 Give positions that come from the main thread priority in the TT. 16MB STC
15-10-13 sni pawn_asymmetry4 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 14563 W: 2826 L: 2636 D: 9101
sprt @ 15+0.05 th 1 Take 4
15-10-13 Fis lazy_easy_fix2 diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 17351 W: 3054 L: 2927 D: 11370
sprt @ 15+0.05 th 3 Easy move fix for lazy smp. This it @mbootsectro's original fix. See commit notes.
15-10-13 Fis lazy_tm_fix diff
ELO: -7.47 +-3.1 (95%) LOS: 0.0%
Total: 20000 W: 3996 L: 4426 D: 11578
20000 @ 5+0.05 th 1 Test for https://github.com/official-stockfish/Stockfish/pull/459
15-10-12 Voy MoveStats diff
ELO: 2.92 +-2.6 (95%) LOS: 98.5%
Total: 20000 W: 3114 L: 2946 D: 13940
20000 @ 60+0.05 th 1 Measure ELO gain of MoveStats patch (LTC)
15-10-07 sni king_separation5 diff
LLR: 1.72 (-2.94,2.94) [0.00,4.00]
Total: 162169 W: 25085 L: 24513 D: 112571
sprt @ 60+0.05 th 1 Try king separation bonus. Take 3257.
15-10-13 sni pawn_asymmetry4 diff
LLR: -0.97 (-2.94,2.94) [0.00,5.00]
Total: 5124 W: 952 L: 971 D: 3201
sprt @ 15+0.05 th 1 Take 3 (rebased, fixed bench)
15-10-13 Voy PT1 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 16547 W: 3075 L: 3175 D: 10297
sprt @ 15+0.05 th 1 I believe these stat's weights are far from optimal.
15-10-12 mco fix_time_bug diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 48089 W: 8851 L: 8777 D: 30461
sprt @ 15+0.05 th 1 Fix time bug once and for all. Idea by Vadim, Gull's author.
15-10-12 IIv reduction_tune diff
19633/20000 iterations
40000/40000 games played
40000 @ 30+0.05 th 1 Tuning my original idea on reduction - retest with carefully chosen ck valus. I'll explain my procedure on the forum, and if this fails, I'll dramatically reduce my using of SPSA.
15-10-12 lbr atomic_signals diff
ELO: 2.85 +-3.3 (95%) LOS: 95.4%
Total: 19661 W: 4640 L: 4479 D: 10542
20000 @ 2+0.02 th 15 Atomic signals: acquire/release on the hot path. 15 threads, 2"+0.02".
15-10-12 lbr atomic_signals^ diff
ELO: 2.59 +-3.4 (95%) LOS: 93.3%
Total: 18132 W: 4113 L: 3978 D: 10041
20000 @ 2+0.02 th 15 Atomic signals: memory_order_seq_cst. 15 threads, 2"+0.02".
15-10-12 lbr atomic_signals diff
ELO: 0.23 +-3.5 (95%) LOS: 55.1%
Total: 18108 W: 4326 L: 4314 D: 9468
20000 @ 2+0.02 th 7 Atomic signals: acquire/release on the hot path. 7 threads, 2"+0.02".
15-10-12 lbr atomic_signals^ diff
ELO: -1.64 +-3.6 (95%) LOS: 18.8%
Total: 16914 W: 4053 L: 4133 D: 8728
20000 @ 2+0.02 th 7 Atomic signals: memory_order_seq_cst. 7 threads, 2"+0.02".
15-10-12 tvi lazy_smp_3 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 4225 W: 662 L: 769 D: 2794
sprt @ 15+0.05 th 3 Only master updates TT for the first 3 ply. Does this make the search more stable?
15-10-12 sni pawn_asymmetry3 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 6798 W: 1220 L: 1317 D: 4261
sprt @ 15+0.05 th 1 Take 2
15-10-12 jos lazy_smp_simple diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 8960 W: 1520 L: 1381 D: 6059
sprt @ 15+0.05 th 7 Simplify altering the search depth of the helper threads. As a side-effect, the old iterative deepening loop logic is being restored. (Same test as simplify_lazy, just created another branch so diff is working properly.)
15-10-09 Voy lazy_smp_cmh diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 36157 W: 5155 L: 4911 D: 26091
sprt @ 60+0.05 th 3 Moving CMH to thread: I believe this may do well at LTC as CMH table takes time to fill up. Would like to get an idea...low throughput.
15-10-12 sni pawn_asymmetry3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6604 W: 1191 L: 1288 D: 4125
sprt @ 15+0.05 th 1 Try pawn asymmetry
15-10-12 IIv reduction_tune diff
309/20000 iterations
631/40000 games played
40000 @ 30+0.05 th 1 Tuning my original idea on reduction - see more under comments.
15-10-11 n_p Tuned_PassedFile diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 33321 W: 6168 L: 6150 D: 21003
sprt @ 15+0.05 th 1 Test SPSA-values of PassedFile.
15-10-11 IIv reduction_check diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 35645 W: 6594 L: 6628 D: 22423
sprt @ 15+0.05 th 1 Trying tuned values for the first 12 moves.
15-10-12 Voy LMRt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9243 W: 1692 L: 1778 D: 5773
sprt @ 15+0.05 th 1 Take 2.
15-10-10 jki MoveStats diff
ELO: 2.02 +-1.3 (95%) LOS: 99.9%
Total: 100459 W: 17344 L: 16761 D: 66354
100000 @ 30+0.05 th 1 Measure ELO gain of MoveStats patch
15-10-11 pec lazy_branching diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 8674 W: 1409 L: 1498 D: 5767
sprt @ 15+0.05 th 7 Correction for time used due to higher branching factor for many threads.
15-10-09 n_p Tune_PassedFile diff
48914/50000 iterations
99546/100000 games played
100000 @ 30+0.05 th 1 Another SPSA-session on PassedFile.
15-10-10 jos simplify_lazy diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 49389 W: 8516 L: 8442 D: 32431
sprt @ 15+0.05 th 3 Simplify altering the search depth of the helper threads. As a side-effect, the old iterative deepening loop logic is being restored. (Both branches now with 'Retire rook contact checks'.)
15-10-11 Fis max_half_time diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14573 W: 2711 L: 2773 D: 9089
sprt @ 15+0.05 th 1 Take 3 on Vadim's idea. Likely final try.
15-10-10 Fis max_half_time diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 44410 W: 8251 L: 8184 D: 27975
sprt @ 15+0.05 th 1 Never use more than half of remaining time. Suggested by Vadim(author of Gull)
15-10-11 sni check diff
LLR: -2.99 (-2.94,2.94) [0.00,5.00]
Total: 24285 W: 4481 L: 4503 D: 15301
sprt @ 15+0.05 th 1 Suppress popcounts in check analysis (take 2: use original values for checking bonus)
15-10-10 IIv reduction_tune diff
9910/10000 iterations
20000/20000 games played
20000 @ 30+0.05 th 1 Tuning moves 7-12, session 3. Read more on the forum.
15-10-10 mco tune_imbalance diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 36636 W: 5575 L: 5617 D: 25444
sprt @ 60+0.05 th 1 LTC: Test imbalance tuning after 75K games
15-10-10 Mys QCC diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 42691 W: 7931 L: 7872 D: 26888
sprt @ 15+0.05 th 1 See how also scoring Queen contact checks inclusive of King performs.
15-10-11 Fis max_half_time diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20821 W: 3857 L: 3892 D: 13072
sprt @ 15+0.05 th 1 1/4 of time bank only. Take 2 on Vadim's idea.
15-10-11 lbr time_loss_fix diff
ELO: -2.43 +-4.3 (95%) LOS: 13.6%
Total: 10000 W: 1991 L: 2061 D: 5948
10000 @ 2+0.01 th 7 I hope to see lots of time losses with lazy_smp here :-)
15-10-10 bin pawn_imbalance diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8998 W: 1629 L: 1716 D: 5653
sprt @ 15+0.05 th 1 try 3 (fixed)
15-10-10 sni check diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3811 W: 669 L: 779 D: 2363
sprt @ 15+0.05 th 1 Suppress popcounts in check analysis
15-10-10 bin pawn_imbalance diff
LLR: -2.16 (-2.94,2.94) [0.00,5.00]
Total: 31200 W: 5854 L: 5810 D: 19536
sprt @ 15+0.05 th 1 Try 3, against Knight/King should be winning
15-10-10 bin pawn_imbalance diff
LLR: -3.46 (-2.94,2.94) [0.00,5.00]
Total: 15976 W: 2923 L: 3001 D: 10052
sprt @ 15+0.05 th 1 Try 2, just minor imbalance
15-10-10 bin pawn_imbalance diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6933 W: 1225 L: 1321 D: 4387
sprt @ 15+0.05 th 1 Bonus for imbalanced pieces and pawn advantage
15-10-08 sni check diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 84377 W: 15693 L: 15453 D: 53231
sprt @ 15+0.05 th 1 Add king support for queen contact checks. The difference should probably be visible only in late queen endgames, so it's difficult to test this in fishtest.
15-10-10 mco tune_imbalance diff
LLR: 2.94 (-2.94,2.94) [0.00,4.00]
Total: 27667 W: 5247 L: 4994 D: 17426
sprt @ 15+0.05 th 1 Test iimbalance tuning after 75K games
15-10-08 mco tune_imbalance diff
37400/50000 iterations
75915/100000 games played
100000 @ 60+0.05 th 1 Tune imbalance at LTC (recently this was a good approach by Stefano)
15-10-09 IIv lazy_smp_log2 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 23041 W: 3903 L: 3983 D: 15155
sprt @ 15+0.05 th 3 Additional depth 0-1-3. If this fails, I'm convinced that we have almost perfect formula, and I hope that work on lazy_smp will be continued.
15-10-09 Voy LMRt diff
LLR: -3.08 (-2.94,2.94) [0.00,5.00]
Total: 41454 W: 7711 L: 7662 D: 26081
sprt @ 15+0.05 th 1 LMR tweak
15-10-09 sg attack_scaling diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8140 W: 1472 L: 1563 D: 5105
sprt @ 15+0.05 th 1 endgame scaling by attacks of stringer side
15-10-09 Mys CR diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 40838 W: 7652 L: 7600 D: 25586
sprt @ 15+0.05 th 1 Ranked checks?
15-10-07 Mys Ck diff
LLR: -4.05 (-2.94,2.94) [0.00,4.00]
Total: 74348 W: 13859 L: 13816 D: 46673
sprt @ 15+0.05 th 1 Bigger first check bonus with smaller subsequent checks.
15-10-09 IIv lazy_smp_log2 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 12018 W: 1992 L: 2108 D: 7918
sprt @ 15+0.05 th 3 Additional depths 0-1-2. Formula suggested by alwey.