Stockfish Testing Queue

Finished - 1741 tests

16-01-16 SC quietValues diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3105 W: 523 L: 636 D: 1946
sprt @ 10+0.1 th 1 When updating stats, also use value of searched quiets. Try to fix undefined bench issue. Take 2.
16-01-16 SC quietValues diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 812 W: 97 L: 219 D: 496
sprt @ 10+0.1 th 1 When updating stats, also use value of searched quiets. Take 3.
16-01-16 SC quietValues diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4131 W: 717 L: 825 D: 2589
sprt @ 10+0.1 th 1 When updating stats, also use value of searched quiets. Try to fix undefined bench issue.
16-01-13 SC LMR_simp diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 7169 W: 1273 L: 1444 D: 4452
sprt @ 10+0.1 th 1 Use same logic for tweak reductions in cut and noncut nodes. I would test it as a simplification, as it makes the code more readable, but approvers please say your word about it.
16-01-10 SC LMR_history diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 41365 W: 7666 L: 7613 D: 26086
sprt @ 10+0.1 th 1 ... and one with a large positive one.
16-01-10 SC LMR_simpDepthMax diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 5651 W: 977 L: 1145 D: 3529
sprt @ 10+0.1 th 1 Try to also remove explicit reduction increase for moves with bad history and compensate with an offset in the new formula. Test this as a simplification against passed LMR_history, as suggested by VOne.
16-01-10 SC LMR_history diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 11155 W: 1572 L: 1416 D: 8167
sprt @ 60+0.6 th 1 Tuned values. LTC.
16-01-10 SC LMR_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 24009 W: 4397 L: 4419 D: 15193
sprt @ 10+0.1 th 1 I probably have chosen the wrong tuning range for the offset. Two last attempts: one with large negative offset ...
16-01-10 SC LMR_history diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 11140 W: 2117 L: 1944 D: 7079
sprt @ 10+0.1 th 1 Tuned values.
16-01-09 SC LMR_history_tune diff
48365/50000 iterations
98082/100000 games played
100000 @ 20+0.2 th 1 LMR_history was a very close call. Try to tune further, as the previous session came almost to the boundary, and add an offset parameter to compensate to new average reduction from history. I will stop and change c if I see that it is not moving. Rescheduling because of wrong options.
16-01-10 SC queenBishop_battery diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 28442 W: 5264 L: 5267 D: 17911
sprt @ 10+0.1 th 1 Decrease king safety if there is a queen-bishop battery pointing to the king. Take 3.
16-01-09 SC queenBishop_battery diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 26957 W: 4994 L: 5003 D: 16960
sprt @ 10+0.1 th 1 Decrease king safety if there is a queen-bishop battery pointing to the king. Take 2.
16-01-09 SC queenBishop_battery diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22190 W: 4088 L: 4118 D: 13984
sprt @ 10+0.1 th 1 Decrease king safety if there is a queen-bishop battery pointing to the king.
16-01-09 SC LMR_history_tune diff
4563/50000 iterations
9261/100000 games played
100000 @ 20+0.2 th 1 LMR_history was a very close call. Try to tune further, as the previous session came almost to the boundary, and add an offset parameter to compensate to new average reduction from history. I will stop and change c if I see that it is not moving.
16-01-07 SC LMR_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 139494 W: 26021 L: 25542 D: 87931
sprt @ 10+0.1 th 1 Tuning results.
16-01-07 SC LMR_history_tune diff
29162/30000 iterations
60000/60000 games played
60000 @ 20+0.2 th 1 As LMR_history was not that bad, tune it. Reschedule with higher ck as it was not moving.
16-01-06 SC LMR_history_tune diff
27555/30000 iterations
56841/60000 games played
60000 @ 10+0.1 th 1 As LMR_history was not that bad, tune it.
16-01-06 SC LMR_history diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 31715 W: 5789 L: 5779 D: 20147
sprt @ 10+0.1 th 1 Scale history dependent reductions with history values. Never increase reductions.
16-01-06 SC LMR_history diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 42301 W: 7844 L: 7787 D: 26670
sprt @ 10+0.1 th 1 Scale history dependent reductions with history values. Mantainers please say if you want it tested as a simplification.
16-01-06 SC LMR_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7303 W: 1318 L: 1412 D: 4573
sprt @ 10+0.1 th 1 Scale history dependent reductions with history values. More aggressive values.
16-01-05 SC LMRmovecount diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 81751 W: 14892 L: 15191 D: 51668
sprt @ 10+0.1 th 1 Also replace see by see_sign. Take 2.
16-01-05 SC LMRmovecount diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 54439 W: 9983 L: 9920 D: 34536
sprt @ 10+0.1 th 1 Remove a mysterious condition from reduction decrease for move escaping a capture. It does not actually change the bench, so test as -3,1
16-01-05 SC LMRmovecount diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 2122 W: 323 L: 485 D: 1314
sprt @ 10+0.1 th 1 Locally, it looked like one could go with LMR also for the first move. As the framework is empty, let me test it, just to be sure that I dont see ghosts.
16-01-05 SC variance3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5152 W: 875 L: 978 D: 3299
sprt @ 10+0.1 th 1 A further attempt at my old variance3 patch, after incorporating hints from local tuning.
16-01-04 SC aspiration_higher diff
LLR: -2.97 (-2.94,2.94) [0.00,4.00]
Total: 10727 W: 1947 L: 2068 D: 6712
sprt @ 10+0.1 th 1 See what happens by increasing size of aspiration window by a good amount. (Local tuning related attempt).
16-01-02 SC razor_simp2 diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 7989 W: 1375 L: 1546 D: 5068
sprt @ 10+0.1 th 1 More or less random tweak on razoring.
16-01-02 SC razor_simp diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 4490 W: 756 L: 898 D: 2836
sprt @ 10+0.1 th 1 Razor at high depth with a logarithmic dependency of margin. This time as a tuning patch as suggested by Joona.
16-01-01 SC assorted_tuning2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 50841 W: 9377 L: 9358 D: 32106
sprt @ 10+0.1 th 1 Combine 3 recent tuning patches which failed yellow at STC and give them a second chance: probcut^ by lucasart (search), Shelter by mstembera (pawn eval) and outpost_rwo by Stefano80 (outpost eval).
15-12-31 SC outpost_rwo diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 73039 W: 13599 L: 13502 D: 45938
sprt @ 10+0.1 th 1 Outpost values tuned with this new RWO optimizer. Let us see.
15-11-23 SC tuneRazor diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 86455 W: 16672 L: 16522 D: 53261
sprt @ 10+0.1 th 1 Take 2
15-11-23 SC tuneRazor diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 13228 W: 2515 L: 2627 D: 8086
sprt @ 10+0.1 th 1 Try to fix non monotonic razor margin.
15-11-23 SC tuneRazor diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 3240 W: 541 L: 688 D: 2011
sprt @ 10+0.1 th 1 Have a go at razoring at higher depths while there.
15-11-22 SC varianceRazor diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4729 W: 838 L: 944 D: 2947
sprt @ 10+0.1 th 1 Last attempt: values tuned locally.
15-11-15 SC varianceRazor diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 10331 W: 1970 L: 2051 D: 6310
sprt @ 10+0.1 th 1 Variance in razoring. To make razor pruning more flat over variance, increase razor margin when variance is low.
15-11-14 SC varianceRazor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6251 W: 1167 L: 1266 D: 3818
sprt @ 10+0.1 th 1 Logging fast games locally showed that razor probability increase for low variance. Try to razor only if variance is less than average.
15-11-11 SC variance3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 20315 W: 3933 L: 3969 D: 12413
sprt @ 10+0.1 th 1 The tuning did not work at all. So go back to manual mode: scale variance contribution such that delta(initial_position) = 18.
15-11-09 SC variance3 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 1003 W: 131 L: 252 D: 620
sprt @ 10+0.1 th 1 I think I have badly screwed my tuning session. But let us test anyway whether I have gone somewhere meaningful.
15-11-08 SC varAspTuning diff
19607/20000 iterations
40000/40000 games played
40000 @ 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. Rescheduled after deltaLinear went to the upper limit.
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.
15-11-07 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.
15-11-02 SC init_tuned diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 42809 W: 8336 L: 8341 D: 26132
sprt @ 10+0.1 th 1 Maybe is worth a further attempt: extrapolate from values which failed yellow at STC. Take 4.
15-11-02 SC variance3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5464 W: 972 L: 1074 D: 3418
sprt @ 10+0.1 th 1 Variance stuff, take 3. Scale aspiration window with variance of root position.
15-11-01 SC init_tuned diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 49900 W: 9650 L: 9630 D: 30620
sprt @ 10+0.1 th 1 Try to extract some useful information from the initiative tuning. Take 3 and then call it a day.
15-11-01 SC variance2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 9759 W: 1858 L: 1941 D: 5960
sprt @ 10+0.1 th 1 Take2 : add variance to futility margin.
15-11-01 SC variance diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8233 W: 1580 L: 1670 D: 4983
sprt @ 10+0.1 th 1 An exotic (probably wrong) idea: interpret one term in evaluate initiative as position variance and avoid tuning at shallow depth if this variance is very high.
15-11-01 SC generalize_init2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 2979 W: 515 L: 629 D: 1835
sprt @ 10+0.1 th 1 Game phase dependent initiative. Take 2.
15-10-31 SC generalize_init diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6831 W: 1310 L: 1407 D: 4114
sprt @ 10+0.1 th 1 Tuned values from quadratic, game phase dependent initiative generalization. Some coefficients have gone very far away from initial values, which makes this very interesting.
15-10-30 SC quadratic_initiative_tu diff
46131/100000 iterations
137678/200000 games played
200000 @ 10+0.1 th 1 Fix base branch. Not my day, in any case.
15-10-30 SC quadratic_initiative_tu diff
10/50000 iterations
16/100000 games played
100000 @ 10+0.1 th 1 Bugfix, thanks to Rocky640