Stockfish Testing Queue

Finished - 24198 tests

08-08-14 ur minimaldepth2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 59146 W: 9883 L: 9844 D: 39419
sprt @ 15+0.05 th 1 use minimal depth for some pruning and also for null move reduction. help to see the following mate in 4 at depth 24 7k/1p3p2/4pNpB/1b2K3/3p4/8/8/8 w - - 0 1
08-08-14 ee probCut_b diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 21220 W: 3506 L: 3558 D: 14156
sprt @ 15+0.05 th 1 Start ProbCut at same depth as nullmove.
08-08-14 ur less_recursive_reductio diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 7282 W: 1480 L: 1567 D: 4235
sprt @ 15+0.05 th 1 testing again and hope this time the test is going to run after fixing the compiler error.
06-08-14 Fi secondary_mobility diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 3736 W: 591 L: 685 D: 2460
sprt @ 15+0.05 th 1 Don't exclude squares attacked by enemy pawns but occupied by enemy pieces from mobility area.
05-08-14 ee probCut diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14737 W: 2396 L: 2464 D: 9877
sprt @ 15+0.05 th 1
05-08-14 Fi unexpected_pos_time diff
ELO: -3.16 +-2.8 (95%) LOS: 1.3%
Total: 19999 W: 3302 L: 3484 D: 13213
20000 @ 15+0.05 th 1 Search longer when opponent makes an unpredicted move. Tuning 4 Low priority
05-08-14 Fi secondary_mobility diff
ELO: -10.32 +-2.8 (95%) LOS: 0.0%
Total: 20000 W: 3161 L: 3755 D: 13084
20000 @ 15+0.05 th 1 Give a partial mobility bonus for squares that are occupied by our pawns or attacked by enemy pawns. Tuning 1
04-08-14 jo insufficient diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 33023 W: 4613 L: 4509 D: 23901
sprt @ 60+0.05 th 1 LTC: Remove check for insufficient material. Seems to kick in rather rare (0.00114%), and draw evaluations for KK, KNK and KBK are covered by material.cpp. Test as simplification.
03-08-14 jo insufficient diff
LLR: 0.05 (-2.94,2.94) [-3.00,1.00]
Total: 128000 W: 21357 L: 21560 D: 85083
sprt @ 15+0.05 th 1 Remove check for insufficient material. Seems to kick in rather rare (0.00114%), and draw evaluations for KK, KNK and KBK are covered by material.cpp. Test as simplification.
04-08-14 ur min_depth1 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 22832 W: 3841 L: 3889 D: 15102
sprt @ 15+0.05 th 1 It is a functional change inspite of having the same bench and I test replacing depth/4 by depth/5(depth/3 seems to me too much minimal depth at least for long time control and if depth/5 pass 15+0.05 then I prefer it to depth/4)
04-08-14 ur minimal_depth diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 88790 W: 14788 L: 14678 D: 59324
sprt @ 15+0.05 th 1 limit the number of plies in searches after null in order to have at least iteration/4 plies(I think to do it also for LMR but I prefer to test first only one change per time(test signature earlier was wrong and this is the reason that the test failed so I correct it)
04-08-14 Fi unexpected_pos_time diff
ELO: -0.61 +-2.8 (95%) LOS: 33.4%
Total: 20000 W: 3302 L: 3337 D: 13361
20000 @ 15+0.05 th 1 Search longer when opponent makes an unpredicted move. Tuning 3 likely final
04-08-14 Fi TTkey_21bit diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 13873 W: 2244 L: 2314 D: 9315
sprt @ 15+0.05 th 1 Change TT entry/cluster code to use 21-bit instead of 16-bit keys to reduce hard collisions by a factor of 32. 4MB hash for STC to simulate real tournament pressure.
04-08-14 Fi unexpected_pos_time diff
ELO: -2.45 +-2.8 (95%) LOS: 4.3%
Total: 20000 W: 3299 L: 3440 D: 13261
20000 @ 15+0.05 th 1 Search longer when opponent makes an unpredicted move. Tuning 2
04-08-14 Fi unexpected_pos_time diff
ELO: -1.54 +-2.1 (95%) LOS: 7.3%
Total: 36387 W: 6047 L: 6208 D: 24132
20000 @ 15+0.05 th 1 Search longer when opponent makes an unpredicted move. Tuning 1
03-08-14 jo insufficient diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 29967 W: 4961 L: 4992 D: 20014
sprt @ 15+0.05 th 1 Take 2. Extend to KBKB (bishops of opposite colors excluded) and KBKN only.
03-08-14 ee ext_verify_e diff
ELO: -10.24 +-2.9 (95%) LOS: 0.0%
Total: 18727 W: 2884 L: 3436 D: 12407
20000 @ 15+0.05 th 1 (I made some comments on #!topic/fishcooking/hFYUD43VtSs)
03-08-14 jo insufficient diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1863 W: 290 L: 390 D: 1183
sprt @ 15+0.05 th 1 Extend test of insufficient material. Though this may give some false positives in very uncommon endgames like KBKB or KBKN, let's see if it helps.
02-08-14 aj remove_unsupported diff
LLR: -2.94 (-2.94,2.94) [-3.00,1.00]
Total: 20705 W: 3504 L: 3693 D: 13508
sprt @ 15+0.05 th 1 test as simplification:remove unsupported pawns: Local tests show ELO gain on removing it : STC
02-08-14 aj remove_unsupported diff
ELO: 1.34 +-2.8 (95%) LOS: 82.1%
Total: 20000 W: 3544 L: 3467 D: 12989
20000 @ 15+0.05 th 1 Compute ELO of unsupported pawns: Local tests show ELO gain on removing it : STC
02-08-14 jo clop-matimb2 diff
ELO: -1.00 +-1.9 (95%) LOS: 14.6%
Total: 49998 W: 9236 L: 9380 D: 31382
50000 @ 7+0.05 th 1 Test first set of CLOP values.
01-08-14 hx kingsafety diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18097 W: 3065 L: 3124 D: 11908
sprt @ 15+0.05 th 1 more safety
01-08-14 hx kingsafety diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 5302 W: 874 L: 964 D: 3464
sprt @ 15+0.05 th 1 less safety
01-08-14 za passed_con diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14612 W: 2431 L: 2499 D: 9682
sprt @ 15+0.05 th 1 Add bonus for passed connected pawns, I think it has same bench due to unlikely case.
30-07-14 jo add_KBBK diff
ELO: -0.42 +-2.5 (95%) LOS: 37.3%
Total: 20000 W: 2739 L: 2763 D: 14498
20000 @ 60+0.05 th 1 Quick check at LTC. Add draw detection for KBBsK with bishops on same color to KXK endgame. Fixed code to not give false positives. First test passed STC, but failed LTC. http://tests.stockfishchess.org/tests/view/535040c50ebc5978b5cf77b0
01-08-14 ee unprotectedPawns diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4596 W: 725 L: 817 D: 3054
sprt @ 15+0.05 th 1 I hope the popcount<>() is not too costly for this.
31-07-14 My Connected_passed2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 15052 W: 2611 L: 2677 D: 9764
sprt @ 15+0.05 th 1 Reward passed connected pawns. Take 2
31-07-14 My Connected_passed diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 12231 W: 1980 L: 2053 D: 8198
sprt @ 15+0.05 th 1 Reward passed connected pawns.
29-07-14 za passed_king diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 23687 W: 3905 L: 4098 D: 15684
sprt @ 15+0.05 th 1 Check the capture square instead of next-next square (simplification).
29-07-14 jo add_KBBK diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 59723 W: 10023 L: 9982 D: 39718
sprt @ 15+0.05 th 1 Add draw detection for KBBsK with bishops on same color to KXK endgame. Fixed code to not give false positives. First test passed STC, but failed LTC. http://tests.stockfishchess.org/tests/view/535040c50ebc5978b5cf77b0
28-07-14 lb norings diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 59814 W: 9681 L: 9929 D: 40204
sprt @ 15+0.05 th 1 no rings
29-07-14 Ac tm_testing2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 39021 W: 6465 L: 6494 D: 26062
sprt @ 15+0.05 th 1
29-07-14 Ac tm_testing3 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 17065 W: 2793 L: 2892 D: 11380
sprt @ 15+0.05 th 1 Time management tuning, 3rd test.
28-07-14 Ac time_manage_testing1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 20210 W: 3338 L: 3427 D: 13445
sprt @ 15+0.05 th 1 Attempt at fine tuning the time management values.
28-07-14 ee ext_verify_d diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 5735 W: 905 L: 994 D: 3836
sprt @ 15+0.05 th 1 More aggressive nullmove pruning. Take 2
28-07-14 ee ext_verify_c diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1471 W: 196 L: 295 D: 980
sprt @ 15+0.05 th 1 More aggressive nullmove pruning. This is with two changes, will split this if successful.
27-07-14 jo limit_null diff
ELO: -0.96 +-2.5 (95%) LOS: 22.9%
Total: 20000 W: 2716 L: 2771 D: 14513
20000 @ 60+0.05 th 1 Check at LTC, too, as STC might be too insensitive. Limit null-move reduction to 8 plies. (R=7) Quick measurement.
27-07-14 za passed_tweak diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 7227 W: 1178 L: 1264 D: 4785
sprt @ 15+0.05 th 1 Tweak some params on passed pawns evaluate (changed many code - SPRT(-1.5, 4.5))
27-07-14 aj outposts_tune_B diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 6273 W: 992 L: 1080 D: 4201
sprt @ 15+0.05 th 1 Tune bishop and knight outposts separately using SPSA: Take B after about 8K iterations: STC
27-07-14 aj outposts_tune_C diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 25372 W: 4234 L: 4276 D: 16862
sprt @ 15+0.05 th 1 tune outposts : Take 2
27-07-14 jo limit_null diff
ELO: -0.89 +-2.8 (95%) LOS: 26.6%
Total: 20000 W: 3311 L: 3362 D: 13327
20000 @ 15+0.05 th 1 Limit null-move reduction to 8 plies. (R=7) Quick measurement.
27-07-14 lb see diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 5772 W: 894 L: 1058 D: 3820
sprt @ 15+0.05 th 1 sort captures by SEE
27-07-14 lb kingdanger diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 43470 W: 5972 L: 5882 D: 31616
sprt @ 60+0.05 th 1 LTC for hxim: kingdanger array removed stc no-regresssion
26-07-14 hx kingdanger diff
ELO: 2.31 +-2.8 (95%) LOS: 94.8%
Total: 20000 W: 3427 L: 3294 D: 13279
20000 @ 15+0.05 th 1 kingdanger array removed measurement
26-07-14 lb aspiration diff
LLR: 3.51 (-2.94,2.94) [0.00,4.00]
Total: 59841 W: 8345 L: 8006 D: 43490
sprt @ 60+0.05 th 1 aspiration: widen slower. LTC anyway, as it was close enough at STC, and could have good scaling (because higher depth should mean more stable scores at the root). Low prio.
26-07-14 hx kingdanger diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 73427 W: 12331 L: 12295 D: 48801
sprt @ 15+0.05 th 1 kingdanger array removed stc no-regresssion
26-07-14 za passed_king diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 7317 W: 1156 L: 1322 D: 4839
sprt @ 15+0.05 th 1 Check the capture square instead of two next squares (simplification).
26-07-14 aj outposts_tune_A diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 17588 W: 2919 L: 3017 D: 11652
sprt @ 15+0.05 th 1 Tune bishop and knight outposts separately using SPSA: Take A after about 4K iterations: STC
24-07-14 jo ext_verify diff
ELO: -1.16 +-1.7 (95%) LOS: 9.0%
Total: 40000 W: 4910 L: 5044 D: 30046
40000 @ 120+0.05 th 1 Double LTC: Check elo impact of a more accurate verification search. Retest at very long tc for scalability issues, as proposed by Marco. Low priority.
26-07-14 ee verification_small_spee diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 101675 W: 16817 L: 17134 D: 67724
sprt @ 15+0.05 th 1 STC, no-regression test. A speed test should be the most important factor here. No functional change.