Stockfish Testing Queue

Finished - 27229 tests

25-01-16 II aspiration diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 47166 W: 8764 L: 8685 D: 29717
sprt @ 10+0.1 th 1 Changing delta after fails, last try.
25-01-16 Vo lmrt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16549 W: 3079 L: 3133 D: 10337
sprt @ 10+0.1 th 1 Take 2..
25-01-16 Ro BishopMobCorner diff
LLR: -1.65 (-2.94,2.94) [0.00,5.00]
Total: 10230 W: 1870 L: 1896 D: 6464
sprt @ 10+0.1 th 1 Only the Bishop, and only the corners.
25-01-16 pe easy diff
ELO: -1.98 +-2.9 (95%) LOS: 9.3%
Total: 20000 W: 3656 L: 3770 D: 12574
20000 @ 10+0.1 th 1 How much would it lose with naive definition of easy move in the same framework
25-01-16 pb tt_quality_classes2 diff
ELO: -5.21 +-8.0 (95%) LOS: 10.0%
Total: 3000 W: 594 L: 639 D: 1767
10000 @ 2+0.02 th 7 Attempt nr. 3 with modified condition.
25-01-16 Ro BishopMobCorner diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 15818 W: 2874 L: 2932 D: 10012
sprt @ 10+0.1 th 1 Just the corners
25-01-16 pb tt_quality_classes2 diff
ELO: -0.71 +-3.8 (95%) LOS: 35.9%
Total: 12724 W: 2572 L: 2598 D: 7554
20000 @ 2+0.02 th 7 Attempt nr. 2 with simplified code. Did local tests diddle me again?
25-01-16 lb connected diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 8774 W: 1636 L: 1764 D: 5374
sprt @ 10+0.1 th 1 tuned on fishtest
24-01-16 Vo lmrt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 61617 W: 11618 L: 11474 D: 38525
sprt @ 10+0.1 th 1 LMR tweak...
25-01-16 jo mcp_pv diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 4809 W: 824 L: 965 D: 3020
sprt @ 10+0.1 th 1 Try to restore old behaviour. Basically, pruning at pv nodes should have been tested as a parameter tuning patch. See also commit notes, please.
25-01-16 SC evaluate_rook_simp diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 14296 W: 2604 L: 2787 D: 8905
sprt @ 10+0.1 th 1 Simplify trapped rook pattern.
25-01-16 SC evaluate_rook_simp diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 11148 W: 2099 L: 2279 D: 6770
sprt @ 10+0.1 th 1 Retire trapped rook pattern. Fixed bench.
24-01-16 SC SPSA_benchmark diff
ELO: -109.62 +-1.9 (95%) LOS: 0.0%
Total: 74904 W: 9631 L: 32510 D: 32763
100000 @ 10+0.1 th 1 Step 2 in SPSA benchmark, see https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/6tvvoAh9xh4 Average QueenMGValue, measure ELO with low throughput
25-01-16 Ro BishopMobCorner diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4776 W: 891 L: 997 D: 2888
sprt @ 10+0.1 th 1 Remove sides squares from Minor mobility area
24-01-16 lb connected diff
43840/40000 iterations
90000/90000 games played
90000 @ 10+0.1 th 1 connected
24-01-16 lb connected diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 10487 W: 1985 L: 2106 D: 6396
sprt @ 10+0.1 th 1 locally tuned at depth=10
24-01-16 Vo HE diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 3839 W: 663 L: 772 D: 2404
sprt @ 10+0.1 th 1 History Extension. See how this goes...
24-01-16 II aspiration diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11464 W: 2076 L: 2152 D: 7236
sprt @ 10+0.1 th 1 Increasing delta after fails. Take 3.
24-01-16 SC LMR_cutscale diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 3355 W: 546 L: 709 D: 2100
sprt @ 10+0.1 th 1 Scale reductions in cut nodes proportionally to history. Couple to history reductions. Test as a simplification.
23-01-16 II nrf_tune diff
19928/20000 iterations
40000/40000 games played
40000 @ 20+0.2 th 1 Restarted after Nicklas' clarification. New TC.
24-01-16 Vo spd2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 43505 W: 8112 L: 8049 D: 27344
sprt @ 10+0.1 th 1 Take 2...
24-01-16 Ro QInitiative diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 13512 W: 2462 L: 2530 D: 8520
sprt @ 10+0.1 th 1 Initiative adjustment when Q on board.
24-01-16 ci tt_prev_cmh diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9472 W: 1717 L: 1802 D: 5953
sprt @ 10+0.1 th 1 Take 2. Removing CMH update on fail low by ttHit.
24-01-16 SC LMR_cutscale diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 13632 W: 2546 L: 2612 D: 8474
sprt @ 10+0.1 th 1 Scale reductions in cut nodes proportionally to history.
24-01-16 Vo app diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 5718 W: 1037 L: 1138 D: 3543
sprt @ 10+0.1 th 1 No point on advance pawn push restriction for SEE early pruning, since it will just get captures.
24-01-16 SC LMR_random diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 3163 W: 537 L: 649 D: 1977
sprt @ 10+0.1 th 1 Add some randomness in deciding whether to go for a full-depth search. Take 1.
24-01-16 SC LMR_noLargeFH diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 2502 W: 415 L: 530 D: 1557
sprt @ 10+0.1 th 1 Do not research if we exceed alpha by a large margin.
24-01-16 SC LMR_nocut diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 2255 W: 356 L: 519 D: 1380
sprt @ 10+0.1 th 1 Dont handle cut nodes differently than others, and compensate by increasing weight of history reductions.
23-01-16 SC SPSA_benchmark diff
ELO: -386.57 +-4.6 (95%) LOS: 0.0%
Total: 38379 W: 1118 L: 32013 D: 5248
100000 @ 10+0.1 th 1 First step in SPSA benchmark, see https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/6tvvoAh9xh4 Low QueenMGValue, measure ELO with low throughput
24-01-16 SC multiStepsLMR diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 23265 W: 4334 L: 4410 D: 14521
sprt @ 10+0.1 th 1 Limit reductions to 5. Test as a parameter tweak. Take 2.
20-01-16 jk distinct_iteration_11 diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 26691 W: 3594 L: 3481 D: 19616
sprt @ 12+0.12 th 21 LTC: distinct_iter11. Make sure that it doesn't regress with high number of threads.
24-01-16 ci tt_prev_cmh diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16785 W: 3114 L: 3167 D: 10504
sprt @ 10+0.1 th 1 For TT Hit, update CMH for nonPv nodes even if the move is missing, or if failing low.
24-01-16 Ro RJR diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11955 W: 2199 L: 2273 D: 7483
sprt @ 10+0.1 th 1 Coordination penalty when no junction points between R-R, or Q-Rs.
23-01-16 Vo spd diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 59456 W: 11062 L: 10930 D: 37464
sprt @ 10+0.1 th 1 Smarter predicted depth...
23-01-16 II aspiration diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 40891 W: 7648 L: 7596 D: 25647
sprt @ 10+0.1 th 1 Increasing delta after fails. Second try.
19-01-16 pe tm diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 132314 W: 17939 L: 17969 D: 96406
sprt @ 60+0.6 th 1 LTC. Simplification
23-01-16 gl mc_pv_take1 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25702 W: 4718 L: 4733 D: 16251
sprt @ 10+0.1 th 1 Use PseudoAttacks instead of more accurate pos.attacks_from(). Final attempt in this series.
23-01-16 gl mc_pv_take1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15672 W: 2825 L: 2883 D: 9964
sprt @ 10+0.1 th 1 Tweak the movecount pruning curve to adjust for 4% of the moves being skipped.
23-01-16 lb tt diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 23010 W: 4196 L: 4274 D: 14540
sprt @ 10+0.1 th 1 tt overwrite
23-01-16 II asp_tune diff
7978/20000 iterations
16048/40000 games played
40000 @ 10+0.1 th 1 Aspiration patch is struggling. I think it's time to finish this with tuning and one last sprt. I tried all logical ideas and time management simplification is a next goal.
23-01-16 Vo qc diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 2329 W: 381 L: 497 D: 1451
sprt @ 10+0.1 th 1 Use Quiet Move Count for reduction...
23-01-16 II nrf_tune diff
6/20000 iterations
10/40000 games played
40000 @ 30+0.3 th 1 Restarted after Nicklas' clarification.
22-01-16 II nrf_tune diff
16830/20000 iterations
33796/40000 games played
40000 @ 30+0.3 th 1 Trying tuning with rk=0.005. This will be my last attempt on reduction formula, introducing move ordering correction (moc). I'm tuning on 30+0.3, but I'll not go above 40K games. rk=0.005 should give an effect of 100K games (with some statistical error). I continue with locally tuned values.
22-01-16 gl mc_pv_take1 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 31698 W: 5834 L: 5823 D: 20041
sprt @ 10+0.1 th 1 Try more accurate, but slightly more expensive pruning. No significant effect on NPS that I can measure.
20-01-16 My BT diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 25206 W: 4135 L: 4021 D: 17050
sprt @ 10+0.1 th 3 In the past I've noticed thread depth to be less effective than one would think for selecting best moves. I'm hoping this tiny simplification might even add strength.
22-01-16 II aspiration diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11636 W: 2129 L: 2204 D: 7303
sprt @ 10+0.1 th 1 Please check the code. Put priority to -1 so far.
22-01-16 Vo mcpt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 2656 W: 453 L: 568 D: 1635
sprt @ 10+0.1 th 1 Move Count Pruning Tweak (fixed bench)
22-01-16 gl mc_pv_take1 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14712 W: 2708 L: 2770 D: 9234
sprt @ 10+0.1 th 1 I will defeat this PV pruning rule at some point :)
22-01-16 pb tt_quality_classes diff
LLR: -1.00 (-2.94,2.94) [0.00,5.00]
Total: 2655 W: 410 L: 441 D: 1804
sprt @ 5+0.1 th 7 Quality classes in TT for Lazy SMP. Since this is again a low-pressure-patch I use old default Hash=128
22-01-16 Vo ct diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 13268 W: 2416 L: 2527 D: 8325
sprt @ 10+0.1 th 1 Tweak Good Capture's SEE threshold...