Stockfish Testing Queue

Finished - 4995 tests

17-06-01 sg king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11022 W: 1953 L: 2031 D: 7038
sprt @ 10+0.1 th 1 50% more king shelter score for king file (with compensation).
17-06-01 sg king_safety diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10136 W: 1801 L: 1883 D: 6452
sprt @ 10+0.1 th 1 Perhaps if the neighbor files are bad the king is stuck on his file, so try the opposite: 25% less king shelter score for king file (with compensation).
17-06-01 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5101 W: 894 L: 998 D: 3209
sprt @ 10+0.1 th 1 Now compensate by scaling down
17-06-01 sg king_safety diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 1943 W: 334 L: 454 D: 1155
sprt @ 10+0.1 th 1 Double king shelter score for king file
17-05-31 sg passed_pawn_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25541 W: 4545 L: 4562 D: 16434
sprt @ 10+0.1 th 1 Additionally ignore pawn attacks. This effects only candidate passers which needs two pushes to become passed.
17-05-30 sg passed_pawn_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 47742 W: 8566 L: 8489 D: 30687
sprt @ 10+0.1 th 1 Ignore pawns for hinder passed pawn bonus. This effects only the new passed pawn breaks.
17-05-30 sg dangerous_pawns diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 13158 W: 2450 L: 2519 D: 8189
sprt @ 10+0.1 th 1 Generalize by removing b/g pawn from the pattern
17-05-30 sg dangerous_pawns diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 10151 W: 1849 L: 1931 D: 6371
sprt @ 10+0.1 th 1 King safety penalty fr dangerous pawn structure (like white pawn on e5 and black pawns on e6, f7 and g7 with black king on the kingside). Inspired by following thread http://talkchess.com/forum/viewtopic.php?t=64111&postdays=0&postorder=asc&topic_view=flat&start=10
17-05-29 sg simpleLMRNewDepth2Tuned diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 16329 W: 2957 L: 3012 D: 10360
sprt @ 10+0.1 th 1 Used logistic regression to get parameters for non captures
17-05-26 sg simpleLMRNewDepth2Tuned diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 62861 W: 11578 L: 11433 D: 39850
sprt @ 10+0.1 th 1 Radical simplified version of SapphireBrand's LMRNewDepth2Tuned2 patch.
17-05-27 sg simpleLMRNewDepth2Tuned diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 33783 W: 6109 L: 6090 D: 21584
sprt @ 10+0.1 th 1 Double weight of retreat moves
17-05-26 sg simpleLMRNewDepth2Tuned diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 16009 W: 2832 L: 2890 D: 10287
sprt @ 10+0.1 th 1 Another radical simplified version.
17-05-24 sg lmr_history2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 35443 W: 4592 L: 4595 D: 26256
sprt @ 60+0.6 th 1 LTC: 23/16 history weight
17-05-24 sg lmr_history2 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 23257 W: 4306 L: 4082 D: 14869
sprt @ 10+0.1 th 1 23/16 history weight
17-05-24 sg lmr_history2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10067 W: 1805 L: 1887 D: 6375
sprt @ 10+0.1 th 1 The tuning shows that most parameters doesn't move significantly. Only history weight shows a clear downward trend. So test further in this direction with 11/8 history weight.
17-05-23 sg tune_lmr_history diff
29344/30000 iterations
59943/60000 games played
60000 @ 10+0.1 th 1 Tuning the coefficients for LMR history starting from my yellow 3/2 history patch (ck=10)
17-05-22 sg lmr_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14299 W: 2555 L: 2619 D: 9125
sprt @ 10+0.1 th 1 Search for the other factors (cmh,fmh,fmh2) also for better weight to perhaps combine later with the 3/2 history weight. Now try 5/4 cmh
17-05-22 sg lmr_history2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 19386 W: 3479 L: 3569 D: 12338
sprt @ 10+0.1 th 1 My 3/2 history patch failed yellow. perhaps not the history but the average decrease (-1660) of ss->history (more reductions) is the key. So lower the correction factor.
17-05-21 sg lmr_history2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 67113 W: 12426 L: 12262 D: 42425
sprt @ 10+0.1 th 1 3/2 history weight
17-05-21 sg lmr_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9226 W: 1613 L: 1699 D: 5914
sprt @ 10+0.1 th 1 Half fmh2 weight
17-05-20 sg lmr_history2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 39509 W: 7210 L: 7231 D: 25068
sprt @ 10+0.1 th 1 Instead for lmr only double history in stats update
17-05-20 sg lmr_history2 diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 32876 W: 6036 L: 6079 D: 20761
sprt @ 10+0.1 th 1 Double history weight with compensation
17-05-20 sg lmr_history2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 23276 W: 4166 L: 4193 D: 14917
sprt @ 10+0.1 th 1 Triple history weight for LMR
17-05-20 sg lmr_history2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 41789 W: 7678 L: 7624 D: 26487
sprt @ 10+0.1 th 1 Double history weight for LMR
17-05-20 sg lmr_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7083 W: 1200 L: 1295 D: 4588
sprt @ 10+0.1 th 1 Double fmh2 weight for LMR
17-05-20 sg lmr_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7316 W: 1258 L: 1352 D: 4706
sprt @ 10+0.1 th 1 Double cmh weight for LMR
17-05-20 sg lmr_history2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12886 W: 2303 L: 2373 D: 8210
sprt @ 10+0.1 th 1 Double fmh weight for LMR
17-05-20 sg assault_ratio3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12263 W: 2192 L: 2265 D: 7806
sprt @ 10+0.1 th 1 use maximum
17-05-19 sg assault_ratio2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 10175 W: 1897 L: 1979 D: 6299
sprt @ 10+0.1 th 1 Compensate to zero the mean of the attacker/defender diff
17-05-18 sg assault_ratio2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14400 W: 2538 L: 2602 D: 9260
sprt @ 10+0.1 th 1 Try another formula which is symmetric between attackers and defenders.
17-05-18 sg king_danger_tweak diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25505 W: 4670 L: 4686 D: 16149
sprt @ 10+0.1 th 1 Do some king danger calculations everytime
17-05-18 sg assault_ratio diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 21382 W: 3893 L: 3927 D: 13562
sprt @ 10+0.1 th 1 Even lower scaling
17-05-18 sg assault_ratio diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4608 W: 791 L: 897 D: 2920
sprt @ 10+0.1 th 1 Try lower scaling
17-05-18 sg assault_ratio diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 2880 W: 469 L: 582 D: 1829
sprt @ 10+0.1 th 1 Scale king danger according to the assault ratio.
17-05-17 sg eval_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20113 W: 3555 L: 3595 D: 12963
sprt @ 10+0.1 th 1 Use beta instead of alpha
17-05-17 sg eval_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 26421 W: 4726 L: 4739 D: 16956
sprt @ 10+0.1 th 1 Add compensation
17-05-17 sg eval_lmr diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14933 W: 2655 L: 2717 D: 9561
sprt @ 10+0.1 th 1 Weight 32 failed yellow so try double up to weight 64
17-05-16 sg eval_lmr diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15164 W: 2715 L: 2776 D: 9673
sprt @ 10+0.1 th 1 Eval based lmr
17-05-16 sg eval_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 61148 W: 11101 L: 10966 D: 39081
sprt @ 10+0.1 th 1 Increase weight to 32.
17-05-16 sg eval_lmr diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25629 W: 4571 L: 4588 D: 16470
sprt @ 10+0.1 th 1 Weight eval diff 8 times higher and clamp to [-20000...20000].
17-05-08 sg master diff
ELO: 20.25 +-1.6 (95%) LOS: 100.0%
Total: 40000 W: 5587 L: 3258 D: 31155
40000 @ 60+0.6 th 1 Regression test until "Linear Protector bonus by distance". This can be compared with the VLTC regression test.
17-05-12 sg qs_nullmove diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 21450 W: 3808 L: 3843 D: 13799
sprt @ 10+0.1 th 1 Do null move search only if static eval > alpha.
17-05-12 sg qs_nullmove diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4553 W: 767 L: 873 D: 2913
sprt @ 10+0.1 th 1 Do null move search only at first level of qsearch.
17-05-12 sg qs_nullmove diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5886 W: 1036 L: 1136 D: 3714
sprt @ 10+0.1 th 1 Use now minimum of null move search and static eval.
17-05-10 sg fix_wrong_tt_hit diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 68203 W: 12394 L: 12354 D: 43455
sprt @ 10+0.1 th 1 Fix the usage of wrong tt hits in search. Read more in repo comment.
17-05-11 sg qs_nullmove diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18415 W: 3325 L: 3372 D: 11718
sprt @ 10+0.1 th 1 Fix bug. Forgot to correct my tempo to get the same eval as static eval if no good opponent move exists. Stop other test
17-05-11 sg qs_nullmove diff
LLR: -1.93 (-2.94,2.94) [0.00,5.00]
Total: 5136 W: 915 L: 975 D: 3246
sprt @ 10+0.1 th 1 Try in qsearch a null move search to get a safer stand pat value (this recognize threats like hanging pieces)
17-05-10 sg quiets2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8435 W: 1493 L: 1582 D: 5360
sprt @ 10+0.1 th 1 Fix problem with triggered assert. See my comment here https://github.com/locutus2/Stockfish/compare/0c1f119...078aac2
17-05-10 sg quiets diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15320 W: 2760 L: 2820 D: 9740
sprt @ 10+0.1 th 1 Exclude quiet checks for negative stats update.
17-05-09 sg passed_pawn_break3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 56109 W: 10123 L: 10010 D: 35976
sprt @ 10+0.1 th 1 Rewrite my last passed pawn break patch (see https://github.com/official-stockfish/Stockfish/pull/1110) and add the case of a pawn break defended by two opponent pawns but also supported by an own pawn.