Stockfish Testing Queue

Finished - 30521 tests

15-11-09 Voy hpTweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13904 W: 2626 L: 2691 D: 8587
sprt @ 10+0.1 th 1 History Pruning Tweak
15-11-08 Ser prev_moves_3fold diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 52353 W: 10499 L: 10133 D: 31721
sprt @ 10+0.1 th 1 Implement a correct 3-fold repetition detection for previous (i.e. before-root) moves.
15-11-08 lbr tuned_tm diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 59445 W: 9114 L: 9088 D: 41243
sprt @ 40+0.4 th 1 LTC: tuned TM
15-11-08 sni asymmetry diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 59795 W: 11603 L: 11547 D: 36645
sprt @ 10+0.1 th 1 Raise asymmetry bonus by 50%
15-11-08 mco no_thread_base diff
LLR: 2.97 (-2.94,2.94) [-5.00,0.00]
Total: 30990 W: 4945 L: 4942 D: 21103
sprt @ 10+0.1 th 7 Test a SMP "non functional change" big simplification. Although non functional even in SMP case Joona suggests to test with SPRT[-5, 0] to avoid some hidden bug with SMP code: take 2
15-11-08 Voy retireCM diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 63103 W: 12014 L: 12286 D: 38803
sprt @ 10+0.1 th 1 See if we can remove Counter Moves.
15-11-09 Roc Alignment diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15918 W: 3064 L: 3120 D: 9734
sprt @ 10+0.1 th 1 When ranking Quiet moves, give a bonus to R, Q and K move which might escape a potential attack. Rebase with latest master
15-11-08 roh helper-thread-depth diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 13168 W: 2148 L: 1976 D: 9044
sprt @ 10+0.1 th 7 Reduce helper thread depth when running out of available time.
15-11-08 jki timeman_log diff
ELO: -2.05 +-3.0 (95%) LOS: 9.4%
Total: 20000 W: 3955 L: 4073 D: 11972
20000 @ 40/10 th 1 Time management, log-tweak, x/y moves, check ELO impact
15-11-08 Roc RealDistance2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25161 W: 4875 L: 4889 D: 15397
sprt @ 10+0.1 th 1 Take 3: Still verify first move by King, but also last one (near the blockSq)
15-11-07 luc lazy_bestmoves diff
LLR: 0.12 (-2.94,2.94) [-3.00,1.00]
Total: 4685 W: 612 L: 612 D: 3461
sprt @ 60+0.4 th 7 LTC: Don't loop over threads, don't look at their depth, only pick thread 1 if it scores better than main thread (minor simplification).
15-11-08 sg outposts diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 22725 W: 4570 L: 4339 D: 13816
sprt @ 10+0.1 th 1 Bonus for possible outposts.
15-11-08 jki timeman_log diff
ELO: -0.21 +-3.0 (95%) LOS: 44.5%
Total: 20000 W: 3799 L: 3811 D: 12390
20000 @ 10+0.1 th 1 Time management, log-tweak, check ELO impact
15-11-08 jki timeman_log diff
LLR: -0.11 (-2.94,2.94) [0.00,5.00]
Total: 52 W: 11 L: 16 D: 25
sprt @ 40/10 th 1 Time management, log-tweak, x/y moves, check ELO impact
15-11-07 SC varAspTuning diff
21546/30000 iterations
43726/60000 games played
60000 @ 10+0.1 th 1 The two last variance attempts were much better than the first three, it seems that it could work, but it is unclear how it has to be parametrized. Try to tune with SPSA.
15-11-08 mco no_thread_base diff
LLR: -0.11 (-2.94,2.94) [-5.00,0.00]
Total: 3118 W: 460 L: 476 D: 2182
sprt @ 10+0.1 th 7 Test a SMP "non functional change" big simplification. Although non functional even in SMP case Joona suggests to test with SPRT[-5, 0] to avoid some hidden bug with SMP code.
15-11-08 Roc RealDistance2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7634 W: 1410 L: 1503 D: 4721
sprt @ 10+0.1 th 1 Take 2. If King can come closer on move 1, test also if it can on move 2.
15-11-08 lbr tuned_tm diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 13449 W: 2740 L: 2605 D: 8104
sprt @ 15+0 th 1 tuned TM: verify that we don't regress without increment
15-11-08 lbr tuned_tm diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 17153 W: 3407 L: 3187 D: 10559
sprt @ 10+0.1 th 1 tuned TM
15-11-07 jos deepest_thread diff
ELO: -20.87 +-11.7 (95%) LOS: 0.0%
Total: 1000 W: 119 L: 179 D: 702
1000 @ 10+0.1 th 15 Take 3. Loop twice over all threads to pick the best. A quick check with a higher number of threads to see, if the additional effort is worth it.
15-11-07 Roc Maneuvers diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10025 W: 1881 L: 1963 D: 6181
sprt @ 10+0.1 th 1 take 2: no penalty for moving twice the same piece when last move was a capture.
15-11-07 Roc RealDistance diff
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 9421 W: 1908 L: 1738 D: 5775
sprt @ 10+0.1 th 1 When evaluating passers, Increase king distance estimate by 1 if king cannot get any closer on next move.
15-11-07 IIv lazy_log_formula diff
ELO: 0.48 +-4.3 (95%) LOS: 58.6%
Total: 8000 W: 1292 L: 1281 D: 5427
8000 @ 10+0.1 th 7 Final check before final sprt test.
15-11-07 luc lazy_bestmoves diff
ELO: -7.99 +-12.0 (95%) LOS: 9.5%
Total: 1000 W: 143 L: 166 D: 691
1000 @ 10+0.1 th 21 Could this make a much larger difference for many threads?
15-11-07 Voy SuperCM diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25723 W: 3781 L: 3808 D: 18134
sprt @ 60+0.4 th 1 SuperCM- I believe this will do substantially better, at LTC. As there will be more hits and the super CM table will have time to fill up. At that moment, over 210k games at STC, it shows an elo gain...
15-11-08 lbr tuned_tm diff
LLR: 0.27 (-2.94,2.94) [0.00,4.00]
Total: 440 W: 95 L: 79 D: 266
sprt @ 10+0.1 th 1 tuned TM
15-11-07 lbr tune diff
38069/40000 iterations
77782/80000 games played
80000 @ 10+0.1 th 1 tume TM
15-11-07 Roc Maneuvers diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8373 W: 1594 L: 1684 D: 5095
sprt @ 10+0.1 th 1 Adjustments on Quiet move scoring based on last move by "side to move"
15-11-06 mco no_thread_base diff
LLR: -0.08 (-2.94,2.94) [-3.00,1.00]
Total: 34000 W: 5447 L: 5504 D: 23049
sprt @ 10+0.1 th 7 Test for regression a SMP simplification: take 2
15-11-05 lbr lazy_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15285 W: 2417 L: 2480 D: 10388
sprt @ 10+0.1 th 7 increase additional depth with depth. given that LMR is asymptotically proportional to log(depth), this seems logical (extra depths are less and less costly).
15-11-07 SC variance3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 27695 W: 5380 L: 5383 D: 16932
sprt @ 10+0.1 th 1 Variance attempts, take4. Slightly decrease baseline delta.
15-11-07 SC variance3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 21929 W: 4282 L: 4310 D: 13337
sprt @ 10+0.1 th 1 Variance attempts, take 5. Decrease delta with increasing variance.
15-11-07 hxi cmtweak diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18164 W: 3466 L: 3512 D: 11186
sprt @ 10+0.1 th 1 don't update counter move if it has bad history
15-11-07 Voy badCM diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 28694 W: 5529 L: 5528 D: 17637
sprt @ 10+0.1 th 1 Don't use counter move if it has bad stats...
15-11-06 IIv lazy_log_formula diff
ELO: -2.02 +-5.5 (95%) LOS: 23.7%
Total: 5000 W: 809 L: 838 D: 3353
5000 @ 10+0.1 th 7 I hope this patch will reveal something and that I'm close to the final tries.
15-11-06 luc lazy_bestmoves diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 30023 W: 4975 L: 4869 D: 20179
sprt @ 10+0.1 th 7 Don't loop over threads, don't look at their depth, only pick thread 1 if it scores better than main thread (minor simplification). Low throughput. Please stop if bogus...
15-11-05 Voy SuperCM diff
LLR: -1.18 (-2.94,2.94) [0.00,5.00]
Total: 234363 W: 45445 L: 44450 D: 144468
sprt @ 10+0.1 th 1 SuperCM- Based off of two consecutive moves. Let see how this goes...
15-11-06 min Mindbreaker1-KingOnPawn diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 33329 W: 6434 L: 6412 D: 20483
sprt @ 10+0.1 th 1 My modified values in Rocky's patch https://github.com/Rocky640/Stockfish/commit/ead0e47f15386d8514b5f9881217c83cb2c7f10a
15-11-06 sg king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10434 W: 2001 L: 2081 D: 6352
sprt @ 10+0.1 th 1 Half unsafe queen contact check weight. Take 3
15-11-06 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 19848 W: 3770 L: 3809 D: 12269
sprt @ 10+0.1 th 1 Double up unsafe queen contact check weight. Take 2
15-11-06 mco deepest_move diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3500 W: 481 L: 589 D: 2430
sprt @ 10+0.1 th 7 Deepest move: take 2
15-11-06 mbo no_control diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 5322 W: 806 L: 968 D: 3548
sprt @ 10+0.1 th 7 Nowadays we selected the best move from the deepest thread, which somewhat counters the idea of adding extra depth to the helpers. This patch removes the depth control from the helper threads. Testing as simplification.
15-11-06 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 55642 W: 10858 L: 10735 D: 34049
sprt @ 10+0.1 th 1 Give penalty also for unsafe queen contact check
15-11-06 mco no_thread_base diff
LLR: -0.11 (-2.94,2.94) [-3.00,1.00]
Total: 684 W: 116 L: 123 D: 445
sprt @ 10+0.1 th 7 Test for regression a SMP simplification
15-11-06 bin lazy_formula diff
ELO: -46.79 +-38.2 (95%) LOS: 0.7%
Total: 127 W: 17 L: 34 D: 76
5000 @ 10+0.1 th 7 Experimental idea: Helper threads run on reduced Depth (not sure what happens when rootDepth goes negative. Early NPS is 20% faster and evens out at high depth)
15-11-06 sni pieces diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 22766 W: 4450 L: 4474 D: 13842
sprt @ 10+0.1 th 1 Reduce material when ahead in endgame (take 2, half bonus)
15-11-06 cib tt-zero-key16 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14010 W: 2639 L: 2704 D: 8667
sprt @ 10+0.1 th 1 Allow retrieving Transposition Table entries with key16 equal to zero. This is more a bug fix than a new feature.
15-11-06 sni pieces diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3564 W: 662 L: 774 D: 2128
sprt @ 10+0.1 th 1 Reduce material when ahead in endgame (take 3, double bonus)
15-11-05 lbr tt diff
LLR: -2.32 (-2.94,2.94) [0.00,4.00]
Total: 11525 W: 1848 L: 1932 D: 7745
sprt @ 10+0.1 th 7 SMP 7 threads: weigh age twice less. strong hash pressure (32MB @ STC, 7 threads)
15-11-05 sni pieces diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 50673 W: 9806 L: 9706 D: 31161
sprt @ 10+0.1 th 1 Reduce material when ahead in endgame