Stockfish Testing Queue

Finished - 37328 tests

15-10-27 Roc RookAttackTweak diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 11962 W: 2265 L: 2381 D: 7316
sprt @ 10+0.1 th 1 Rook contact checks have been recently removed with convincing success. I think it is because the factor for "undefended" in the attackUnits calculation, have increased from 19 (in SF6) to 27 (current master). This "hand-tuned" test increase this a little bit more to 30, and decrease the QCC accordingly.
15-10-27 SC moreTuned diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 7438 W: 1381 L: 1513 D: 4544
sprt @ 10+0.1 th 1 Results from SPSA tuning session. Asymmetries in the pawn evaluation are interesting, as well new weights for cmh vs h in move ordering.
15-10-20 SC moreTuning diff
60500/150000 iterations
123740/300000 games played
300000 @ 40+0.4 th 1 As I was twice lucky with tuning, let me give a third go with other rarely considered parameters. Framework is mostly idle and a lot of machines want to work. Low throughput, so it uses only time when fishtest is idle.
15-10-27 Voy SWt5 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 13093 W: 2414 L: 2526 D: 8153
sprt @ 10+0.1 th 1 Try 5. Slowly but surely I am finding the optimized values.
15-10-26 Roc FollowUpThreatOnMajor diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17516 W: 3450 L: 3498 D: 10568
sprt @ 10+0.1 th 1 Consider follow-up threats only on higher pieces
15-10-27 lbr simple diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 15134 W: 2814 L: 3000 D: 9320
sprt @ 10+0.1 th 1 take 2
15-10-25 Roc ThreatTuneTest diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 82628 W: 16045 L: 15907 D: 50676
sprt @ 10+0.1 th 1 Testing locally tuned values
15-10-27 lbr simple diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 4433 W: 783 L: 950 D: 2700
sprt @ 10+0.1 th 1 simplify after 8fd34d77
15-10-26 sni threats8 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 133016 W: 25836 L: 25518 D: 81662
sprt @ 10+0.1 th 1 Try delta=S(-3,-3) (after inverse parabolic interpolation of the first two runs)
15-10-27 jhe time_cleanup diff
ELO: -2.57 +-6.0 (95%) LOS: 20.2%
Total: 5000 W: 968 L: 1005 D: 3027
5000 @ 10+0.1 th 1 Quick sanity check.
15-10-26 jos time_inc_fix diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 106048 W: 20190 L: 20220 D: 65638
sprt @ 10+0.1 th 1 Never use more than 90% of increment, when playing with increment. This obviously only has effect when playing with increment like in TCEC! Test as no regression.
15-10-26 lbr master diff
ELO: 61.76 +-1.9 (95%) LOS: 100.0%
Total: 40000 W: 10233 L: 3197 D: 26570
40000 @ 60+0.05 th 1 Regression test, rescheduled correctly, to be able to compare to previous ones: 1/ 8moves book 2/ 60+0.05
15-10-27 pec play_on_increment diff
LLR: 3.33 (-2.94,2.94) [-3.00,1.00]
Total: 49343 W: 9449 L: 9358 D: 30536
sprt @ 10+0.1 th 1 No regression an STC; logically it is still should be gain albeit smaller, but proof of no regression at STC test should be enough. As SF lost again on time in TCEC test old elo gaining solution to playing on increment problem http://tests.stockfishchess.org/tests/view/52df0b250ebc59025698f83b
15-10-27 Voy SWt4 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 29026 W: 5654 L: 5651 D: 17721
sprt @ 10+0.1 th 1 Triple the weights.
15-10-27 aji lazy_log_formula_b diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 17704 W: 2489 L: 2364 D: 12851
sprt @ 60+0.4 th 3 Check for no-regression at 3 threads:LTC
15-10-27 Fis NearDraw2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10708 W: 2062 L: 2141 D: 6505
sprt @ 10+0.1 th 1 See github comments. Final try.
15-10-27 sni islands diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10932 W: 2075 L: 2153 D: 6704
sprt @ 10+0.1 th 1 Take 4: use both pawn holes and pawn span
15-10-27 pec play_on_increment diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 6772 W: 1204 L: 1054 D: 4514
sprt @ 1+1 th 1 LTC. As SF lost again on time in TCEC test old elo gaining solution to playing on increment problem http://tests.stockfishchess.org/tests/view/52df0b250ebc59025698f83b It may still work
15-10-26 Voy SWt3 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 83131 W: 16135 L: 15996 D: 51000
sprt @ 10+0.1 th 1 Simply double the weights. Fixed bench.
15-10-27 Roc QueenThreats2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 24758 W: 4740 L: 4757 D: 15261
sprt @ 10+0.1 th 1 Take 3 on Ajith idea to restore Q threats but only on hanging pieces.
15-10-26 sni islands diff
LLR: -2.99 (-2.94,2.94) [0.00,5.00]
Total: 10836 W: 2039 L: 2119 D: 6678
sprt @ 10+0.1 th 1 Use pawn islands, take 3 (count real islands, not holes)
15-10-27 pec play_on_increment diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 8816 W: 1961 L: 1788 D: 5067
sprt @ 0.25+0.25 th 1 As SF lost again on time in TCEC test old elo gaining solution to playing on increment problem http://tests.stockfishchess.org/tests/view/52df0b250ebc59025698f83b It may still work
15-10-26 IIv maxTime diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8509 W: 1644 L: 1733 D: 5132
sprt @ 10+0.1 th 1 My last try on controlling a maximumTime, Take3.
15-10-26 Roc QueenThreats2 diff
ELO: -0.02 +-3.5 (95%) LOS: 49.5%
Total: 14821 W: 2897 L: 2898 D: 9026
20000 @ 10+0.1 th 1 A variation on Ajith request. Only if Q is not attacked, compute Q attacks on hanging pieces.
15-10-26 sni islands diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 47132 W: 9162 L: 9078 D: 28892
sprt @ 10+0.1 th 1 Use pawn islands, take 2
15-10-26 IIv maxTime diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14693 W: 2733 L: 2795 D: 9165
sprt @ 10+0.1 th 1 My last try on controlling a maximumTime, Take2.
15-10-26 jos time_inc_fix diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 9629 W: 1606 L: 1468 D: 6555
sprt @ 1+2 th 1 Never use more than 90% of increment, when playing with increment. This obviously only has effect when playing with increment like in TCEC! Test as no regression. (with big increment)
15-10-26 IIv maxTime diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 33391 W: 6405 L: 6383 D: 20603
sprt @ 10+0.1 th 1 My last try on controlling the maximumTime (something was wrong with the previous run).
15-10-26 Fis max_half_time diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 24184 W: 4566 L: 4768 D: 14850
sprt @ 10+0.1 th 1 Fix time losses by never using more than half of remaining time. Suggested by Vadim(author of Gull)
15-10-26 Roc QueenThreats2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9017 W: 1688 L: 1775 D: 5554
sprt @ 10+0.1 th 1 Test requested by Ajith: restore Q threats on hanging
15-10-26 aji lazy_log_formula_b diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 40457 W: 7148 L: 7060 D: 26249
sprt @ 10+0.1 th 3 Check for no-regression at 3 threads STC
15-10-26 sni islands diff
LLR: -1.86 (-2.94,2.94) [0.00,5.00]
Total: 9531 W: 1842 L: 1879 D: 5810
sprt @ 10+0.1 th 1 Try pawn islands
15-10-26 sg passed_pawns diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15918 W: 3052 L: 3108 D: 9758
sprt @ 10+0.1 th 1 Rework safe squares bonus (including consider own pieces blocking queening path)
15-10-26 jhe log_depth diff
LLR: 0.87 (-2.94,2.94) [0.00,4.00]
Total: 3000 W: 475 L: 422 D: 2103
sprt @ 10+0.1 th 24 Try to optimize thread depth calculation.
15-10-26 mco master diff
ELO: 79.28 +-4.6 (95%) LOS: 100.0%
Total: 8827 W: 2834 L: 854 D: 5139
40000 @ 60+0.4 th 1 Regression test before TCEC superfinal
15-10-26 Roc FollowUpThreat diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14070 W: 2672 L: 2736 D: 8662
sprt @ 10+0.1 th 1 FUT_20151023_2
15-10-26 tvi SEE diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13418 W: 2580 L: 2647 D: 8191
sprt @ 10+0.1 th 1 Adapt for last pawn. Based on idea of VoyagerOne. Fixed bench
15-10-26 SC log_formula_general diff
ELO: -214.05 +-103.1 (95%) LOS: 0.0%
Total: 31 W: 1 L: 18 D: 12
10000 @ 10+0.1 th 3 Try a threads independent generalization of the log formulas I have seen, based on my excel table https://docs.google.com/spreadsheets/d/1Ub3YsFdK_40Cp0DuoVhTVwbZnJXWfXgFwmxb5CN0gK4/edit#gid=0 Check whether it is a obvious regression on 3 threads (which is the thread count for which the change is largest).
15-10-25 Voy FL2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 68431 W: 13358 L: 13178 D: 41895
sprt @ 10+0.1 th 1 Take 2
15-10-26 Voy SWt2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 28998 W: 5660 L: 5714 D: 17624
sprt @ 10+0.1 th 1 take 2
15-10-26 sni threats8 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 34816 W: 6735 L: 6769 D: 21312
sprt @ 10+0.1 th 1 Decrease threats on queen
15-10-26 Voy SWt diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 34543 W: 6712 L: 6747 D: 21084
sprt @ 10+0.1 th 1 Stats Weight Tweak
15-10-23 aji lazy_log_formula_b diff
ELO: 3.44 +-3.8 (95%) LOS: 96.1%
Total: 10000 W: 1627 L: 1528 D: 6845
10000 @ 10+0.1 th 24 Slightly tweak log formula for 24 cores(with TCEC in mind): STC Current formula is not ideal for 24 cores as it puts more cores at Additional depth 8 than at Additional depth 9(see comments)
15-10-26 Roc TestSEE diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 25966 W: 5027 L: 5093 D: 15846
sprt @ 10+0.1 th 1 This idea was tested as simplification some time ago. But to be accepted, it was asked to run SPRT[0,4]. Now boldly retry this same "butcher" idea by also removing Q threats in SEE. Ref: https://github.com/official-stockfish/Stockfish/pull/366 (last note by Zamar) and https://github.com/official-stockfish/Stockfish/pull/365
15-10-25 sni threats8 diff
LLR: -3.05 (-2.94,2.94) [0.00,4.00]
Total: 23162 W: 4442 L: 4523 D: 14197
sprt @ 10+0.1 th 1 Increase threats on queen
15-10-25 IIv lazy_log_formula diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 10522 W: 1855 L: 1976 D: 6691
sprt @ 10+0.1 th 3 If this will not be at least positive, this is my last try on 3 threads. AdditionalDepths 0-3-4.
15-10-25 Roc MinorSnipers diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 26705 W: 5109 L: 5117 D: 16479
sprt @ 10+0.1 th 1 MS_20151023_3
15-10-25 pec tm_threadcount diff
LLR: -1.66 (-2.94,2.94) [0.00,5.00]
Total: 2258 W: 346 L: 406 D: 1506
sprt @ 10+0.1 th 7 stop search when half of helper threads finished their iteration and best move did not change
15-10-25 Roc MinorSnipers diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12566 W: 2398 L: 2469 D: 7699
sprt @ 10+0.1 th 1 MS_20151023_2
15-10-24 Voy YellowComboReg diff
ELO: 0.40 +-3.2 (95%) LOS: 59.8%
Total: 12955 W: 1825 L: 1810 D: 9320
10000 @ 120+0.1 th 1 Make sure my Yellow Combo patch doesn't regress at XLTC.