Stockfish Testing Queue

Finished - 45450 tests

14-03-15 rst capt_ord diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 13281 W: 2443 L: 2513 D: 8325
sprt @ 15+0.05 th 1 Capture move ordering. Take 6
14-03-15 jos closed_bishop diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7171 W: 1246 L: 1332 D: 4593
sprt @ 15+0.05 th 1 Small additional penalty for bishops in closed positions.
14-03-15 rst capt_ord diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 12805 W: 2296 L: 2368 D: 8141
sprt @ 15+0.05 th 1 Capture move ordering. Take 7.
14-03-15 tzi threat_eval_accurate diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 20750 W: 3918 L: 3969 D: 12863
sprt @ 15+0.05 th 1 more accurate piece threat evaluation without much loss of speed
14-03-15 mco asp_tuning^^ diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 13005 W: 2456 L: 2317 D: 8232
sprt @ 15+0.05 th 1 Aspiration test 14 vs 16
14-03-15 mco asp_tuning^ diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 60708 W: 11285 L: 11023 D: 38400
sprt @ 15+0.05 th 1 Aspiration test 12 vs 16
14-03-15 mco asp_tuning diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 19836 W: 3633 L: 3687 D: 12516
sprt @ 15+0.05 th 1 Aspiration test 10 vs 16
14-03-15 mco master diff
LLR: 0.57 (-2.94,2.94) [-3.00,1.00]
Total: 16451 W: 2623 L: 2619 D: 11209
sprt @ 15+0.05 th 8 Test for no-regression remove of Max Threads per Split Point restriction. This is a special test that requires 8 threads at least and so is set at high priority because available machines are very few anyhow and will not affect other tests.
14-03-15 gli zug diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 2535 W: 387 L: 484 D: 1664
sprt @ 15+0.05 th 1 Don't allow null move if less than 2 moves available to non-pawn pieces
14-03-15 mco master diff
LLR: 1.25 (-2.94,2.94) [-3.00,1.00]
Total: 39757 W: 6540 L: 6538 D: 26679
sprt @ 15+0.05 th 3 Test with no-regression and 3 threads "Split also if no slaves are found"
14-03-16 Fis delta_change diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 22425 W: 4126 L: 4173 D: 14126
sprt @ 15+0.05 th 1 Increase delta a bit slower on fail high/low. Take 3
14-03-16 jos cutNode1 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 14672 W: 2745 L: 2812 D: 9115
sprt @ 15+0.05 th 1 Reduces hit rate in full depth PV search by 1%.
14-03-16 jos closed_king diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 5038 W: 872 L: 964 D: 3202
sprt @ 15+0.05 th 1 Funny idea from A. Shashin's book. This finishes my series of tries to improve play in closed positions. Maybe someone else will find some inspirations. ;-)
14-03-16 mco asp_tuning^^ diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 6191 W: 910 L: 983 D: 4298
sprt @ 60+0.05 th 1 LTC: Aspiration test 14 vs 16
14-03-16 sni autocorrect_CUT_ALL_sta diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 1796 W: 266 L: 365 D: 1165
sprt @ 15+0.05 th 1 take 4
14-03-16 sni autocorrect_CUT_ALL_sta diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 904 W: 107 L: 208 D: 589
sprt @ 15+0.05 th 1 take 5
14-03-16 Mys rBeta diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 17800 W: 3299 L: 3358 D: 11143
sprt @ 15+0.05 th 1 Use a fixed value subtraction for rBeta in singular extension search.
14-03-17 sto quadratic_gamephase diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 11653 W: 2126 L: 2201 D: 7326
sprt @ 15+0.05 th 1 Use a 2nd order polynomial to interpolate through the game phases limits instead of a straight line. There's a free parameter b to tune. Here I choose b=0.013. Take 1.
14-03-17 sni autocorrect_CUT_ALL_sta diff
LLR: -1.31 (-2.94,2.94) [-1.50,4.50]
Total: 39521 W: 7402 L: 7345 D: 24774
sprt @ 15+0.05 th 1 take 7
14-03-17 sni noise_of_sf140314 diff
ELO: 9.90 +-8.9 (95%) LOS: 98.5%
Total: 2000 W: 373 L: 316 D: 1311
2000 @ 15+0.05 th 1 measuring the noise of sf140314 against itself (2000@15'')
14-03-17 sni noise_of_sf140314 diff
ELO: 3.13 +-8.3 (95%) LOS: 77.0%
Total: 2000 W: 306 L: 288 D: 1406
2000 @ 60+0.05 th 1 measuring the noise of sf140314 against itself (2000@60'')
14-03-17 mco asp_tuning^ diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 4471 W: 660 L: 741 D: 3070
sprt @ 60+0.05 th 1 LTC: Aspiration test 12 vs 16
14-03-17 hwi futility_margin_tune_2 diff
ELO: 2.74 +-2.9 (95%) LOS: 96.7%
Total: 20000 W: 3769 L: 3611 D: 12620
20000 @ 15+0.05 th 1 Another check on tuning futilty margin
14-03-17 rst asp_start diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 17331 W: 2675 L: 2698 D: 11958
sprt @ 60+0.05 th 1 LTC: aspiration window start size based on previous score
14-03-17 mco bigger_asp diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 34039 W: 5252 L: 5200 D: 23587
sprt @ 60+0.05 th 1 Increase aspiration to 18. Directly at LTC because it seems to be TC dependent (see previous asp. patches).
14-03-17 sto quadratic_gamephase diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 14527 W: 2596 L: 2663 D: 9268
sprt @ 15+0.05 th 1 2nd order polynomial for game phase instead of straight line, with b=0.009. Take 2 (Final take if it fails).
14-03-17 Fis asp_delta_change diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 22725 W: 4091 L: 4137 D: 14497
sprt @ 15+0.05 th 1 Update deltas for alpha and beta independently.
14-03-18 pec still_first diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 13231 W: 1991 L: 2033 D: 9207
sprt @ 60+0.05 th 1 Direct LTC as looks tc dependent . Chip away more time from unchanging moves. Final attempt.
14-03-18 mco split_moves diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 16972 W: 2757 L: 2938 D: 11277
sprt @ 15+0.05 th 3 Test "don't allocate a thread if no moves" for no-regression, this is not enough to commit, but at least check patch is not broken. (3 threads used)
14-03-18 mco split_moves diff
ELO: -10.83 +-7.3 (95%) LOS: 0.2%
Total: 2567 W: 344 L: 424 D: 1799
20000 @ 15+0.05 th 16 Hit the bullet and test split_moves for a gain: we need to use at least 16 threads!
14-03-18 sni LMR_captures diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2304 W: 388 L: 488 D: 1428
sprt @ 15+0.05 th 1 Use LMR for captures too
14-03-18 Mys attack_pin diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 8444 W: 1523 L: 1606 D: 5315
sprt @ 15+0.05 th 1 Give bonus to attacking pins.
14-03-18 Fis asp_interpolate diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 42536 W: 7868 L: 7862 D: 26806
sprt @ 15+0.05 th 1 Interpolate aspiration over a certain depth range.
14-03-18 rst lmr_cap_qsearch diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4284 W: 758 L: 852 D: 2674
sprt @ 15+0.05 th 1 lmr captures on qsearch fail low
14-03-18 Mys tune_asp_del diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 36648 W: 6806 L: 6815 D: 23027
sprt @ 15+0.05 th 1 First and last try at tuning. Increase depth & delta by 2.
14-03-18 sni LMR_captures diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4159 W: 730 L: 824 D: 2605
sprt @ 15+0.05 th 1 LMR : do it for captures too & retire cutNode difference
14-03-19 Fis asp_interpolate diff
LLR: -2.94 (-2.94,2.94) [0.00,6.00]
Total: 65148 W: 10066 L: 9875 D: 45207
sprt @ 60+0.05 th 1 Interpolate aspiration over a certain depth range. It failed STC w/ a neutral score but we know this is TC dependent. I would like to try LTC anyway w/ low priority. If don't agree don't approve.
14-03-19 Fis asp_previous diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 6087 W: 1097 L: 1186 D: 3804
sprt @ 15+0.05 th 1 Bias delta based on delta from previous iteration. Take 1
14-03-19 Fis asp_previous diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 3249 W: 548 L: 644 D: 2057
sprt @ 15+0.05 th 1 Bias delta based on delta from previous iteration. Take 2
14-03-19 sni LMR_captures diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 7858 W: 1422 L: 1507 D: 4929
sprt @ 15+0.05 th 1 LMR for captures, with reduction 3 plies less
14-03-19 sg rook_ending diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 12410 W: 2233 L: 2306 D: 7871
sprt @ 15+0.05 th 1 drawish rook ending (Take 1)
14-03-19 sg rook_ending diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 20497 W: 3759 L: 3811 D: 12927
sprt @ 15+0.05 th 1 drawish rook ending (Take 2)
14-03-19 luc asp_start_retry diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 40265 W: 7322 L: 7323 D: 25620
sprt @ 15+0.05 th 1 aspiration window start size based on previous score: Ralph's patch with very different values loosely based on local testing (while the framework is empty)
14-03-19 sg rook_ending diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 17126 W: 3191 L: 3251 D: 10684
sprt @ 15+0.05 th 1 drawish rook ending (Take 3)
14-03-19 sg rook_ending diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 36603 W: 6794 L: 6804 D: 23005
sprt @ 15+0.05 th 1 drawish rook ending (Take 4)
14-03-20 sni probcut_tweak diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 8799 W: 1568 L: 1650 D: 5581
sprt @ 15+0.05 th 1 Harden probcut precondition, and reduction = 4.5 plies
14-03-20 sni probcut_tweak diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 36543 W: 6869 L: 6878 D: 22796
sprt @ 15+0.05 th 1 Harden probcut precondition, and reduction = 4 plies (current master)
14-03-20 Mys rBeta_tweak diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 9889 W: 1883 L: 1752 D: 6254
sprt @ 15+0.05 th 1 Introduce a factor of depth / 4 to the singular extension margin.
14-03-20 Mys rBeta_tweak diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 16402 W: 2536 L: 2563 D: 11303
sprt @ 60+0.05 th 1 LTC: Introduce a factor of depth / 4 to the singular extension margin.
14-03-20 jos iid_tweak^ diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 21029 W: 3838 L: 3888 D: 13303
sprt @ 15+0.05 th 1 As we arrive here with under 1% as PvNode && !ttMove, I simplified the precondition. As this noticeably affects functionality, I test in standard mode. Take 1.