Stockfish Testing Queue

Finished - 46859 tests

15-03-14 sg more_lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 10786 W: 1981 L: 2101 D: 6704
sprt @ 15+0.05 th 1 Now with better move ordering try a bit more agressive LMR
15-03-14 jki smpSimple diff
LLR: -2.95 (-2.94,2.94) [-3.50,0.50]
Total: 11550 W: 1704 L: 1882 D: 7964
sprt @ 30+0.05 th 4 Remove a split condition which only affects Threads == 4 case.
15-03-14 jki yieldlock diff
ELO: 2.00 +-2.7 (95%) LOS: 92.4%
Total: 20000 W: 3276 L: 3161 D: 13563
20000 @ 15+0.05 th 7 Fully yielding locks
15-03-15 sg more_lmr diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 29074 W: 5767 L: 5503 D: 17804
sprt @ 15+0.05 th 1 Now with better move ordering try a bit more agressive LMR (Take 2)
15-03-15 sg spsa_lmr diff
25619/25000 iterations
40789/50000 games played
50000 @ 15+0.05 th 1 The last test passed STC. So try a tuning session starting from this values to squeeze out more elo.
15-03-15 vin blocker3 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7185 W: 1339 L: 1426 D: 4420
sprt @ 15+0.05 th 1 Try assaying the bonus against any blocker when the pawn is supported from behind (since this can root the blocker to the spot)
15-03-15 lbr history diff
ELO: -11.08 +-8.9 (95%) LOS: 0.7%
Total: 2071 W: 333 L: 399 D: 1339
20000 @ 15+0.05 th 3 smp friendly history 3 threads
15-03-15 Fis time_tuning2 diff
43081/37500 iterations
80000/75000 games played
75000 @ 15+0.05 th 1 Tune some time parameters now that fast move is in. Pri -1
15-03-15 jki thread_clean diff
ELO: -8.92 +-3.1 (95%) LOS: 0.0%
Total: 15394 W: 2265 L: 2660 D: 10469
20000 @ 15+0.05 th 7 Retire native win32 mutexes and condition variables (7 threads)
15-03-15 jki latejoin_ignore diff
ELO: -2.78 +-2.7 (95%) LOS: 2.2%
Total: 20000 W: 3056 L: 3216 D: 13728
20000 @ 15+0.05 th 16 Ignore MAX_SLAVES_PER_SPLITPOINT when latejoining, 16 threads
15-03-15 jki helpful diff
LLR: -2.95 (-2.94,2.94) [-1.50,6.50]
Total: 5020 W: 738 L: 795 D: 3487
sprt @ 15+0.05 th 16 Final run to determine if orthodox helpful master concept is a minor boost with 16 threads or not. Exceptional SPRT bounds (-1.5, 6.5) used.
15-03-15 sni pawn_mobility diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 5824 W: 1106 L: 1197 D: 3521
sprt @ 15+0.05 th 1 Raise value of pawn mobility.
15-03-16 sg more_lmr diff
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 26674 W: 5278 L: 5023 D: 16373
sprt @ 15+0.05 th 1 The tuning changes seems very small (but this parameters a vey sensitive to changes), so lets test it.
15-03-16 mco spinlock_fix diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 25928 W: 4303 L: 4190 D: 17435
sprt @ 15+0.05 th 4 Regression test for an important bug fix in new spinlock code
15-03-16 sg more_lmr diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 129049 W: 20590 L: 20346 D: 88113
sprt @ 60+0.05 th 1 LTC: The tuning changes seems very small (but this parameters a vey sensitive to changes), so lets test it.
15-03-16 Roc BackwardSimplification diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 42455 W: 8199 L: 8206 D: 26050
sprt @ 15+0.05 th 1 Handle backward penalty exactly the same as if it was an isolated pawn
15-03-16 lan bishop_colored_pawns diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 8294 W: 1523 L: 1607 D: 5164
sprt @ 15+0.05 th 1 Penalty for pawns of the same color as bishop. Suggested by 'dragon'
15-03-16 vin stranded_pawn diff
LLR: 0.49 (-2.94,2.94) [-1.50,4.50]
Total: 5540 W: 1130 L: 1097 D: 3313
sprt @ 15+0.05 th 1 Inspired by a Mindbreaker post - treat stranded pawns like isolated ones.
15-03-16 mco nosleep_pull diff
ELO: -1.32 +-3.9 (95%) LOS: 25.6%
Total: 10000 W: 1653 L: 1691 D: 6656
10000 @ 15+0.05 th 4 Quick test to verify SMP cleanup patch
15-03-17 vin stranded_pawn diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 34619 W: 6659 L: 6672 D: 21288
sprt @ 15+0.05 th 1 Inspired by a Mindbreaker post - treat stranded pawns like isolated ones. This is the same algorithm/bench sig but with the windows crash bug fixed.
15-03-18 vin stranded_pawn diff
LLR: -4.99 (-2.94,2.94) [-1.50,4.50]
Total: 16110 W: 3108 L: 3244 D: 9758
sprt @ 15+0.05 th 1 Take 2. Try just half the full isolated pawn penalty.
15-03-18 mco thread_clean diff
ELO: -16.66 +-14.2 (95%) LOS: 1.1%
Total: 668 W: 82 L: 114 D: 472
20000 @ 15+0.05 th 7 Respin oj Joona's "Retire native win32 mutexes and condition variables (7 threads)". Now with mutex bug fixed.
15-03-18 lan bishop_colored_pawns diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7825 W: 1468 L: 1553 D: 4804
sprt @ 15+0.05 th 1 Take 2. Penalties for pawns on the same-colored square as bishop.
15-03-18 jos mobility2 diff
ELO: 3.40 +-2.2 (95%) LOS: 99.9%
Total: 40000 W: 8434 L: 8042 D: 23524
40000 @ 9+0.05 th 1 Let's check if there is some further improvement after another SPSA run.
15-03-18 vin stranded_pawn diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 20213 W: 3844 L: 3896 D: 12473
sprt @ 15+0.05 th 1 Last try. Penalise the stranded pawn as a double unsupported penalty. (Note that without this code, the pawn would have been classed as unsupported).
15-03-18 Roc BackwardSimplification diff
LLR: 3.65 (-2.94,2.94) [-3.00,1.00]
Total: 41601 W: 8131 L: 8009 D: 25461
sprt @ 15+0.05 th 1 Handle backward penalty exactly the same as if it was an isolated pawn. Rescheduled for "simplification". See discussion on https://github.com/Rocky640/Stockfish/commit/5d1071da0e57290baa67b9bfe90c0543682ded1b
15-03-18 mco thread_clean diff
ELO: -0.66 +-2.7 (95%) LOS: 31.8%
Total: 20000 W: 3210 L: 3248 D: 13542
20000 @ 15+0.05 th 7 Use std::condition_variable_any instead of our own version
15-03-18 Fis tunedTime2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 14890 W: 2858 L: 2924 D: 9108
sprt @ 15+0.05 th 1 SPSA tuned time values.
15-03-19 vin doubled_passer_2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 4409 W: 766 L: 860 D: 2783
sprt @ 15+0.05 th 1 Try this idea with Mindbreaker's suggested values.
15-03-19 jos mobility2 diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 11052 W: 1874 L: 1687 D: 7491
sprt @ 60+0.05 th 1 LTC: Further improved mobility values.
15-03-19 sg fmh diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 5236 W: 925 L: 1016 D: 3295
sprt @ 15+0.05 th 1 Followup moves history. Variation of hxim's idea.
15-03-19 Roc LeverAdjust diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 5372 W: 979 L: 1071 D: 3322
sprt @ 15+0.05 th 1 When isolated or unsupported are attacked, lower the penalty. From a discussion with CuddlestMonkey
15-03-19 Roc BackwardSimplification diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 206444 W: 32816 L: 32982 D: 140646
sprt @ 60+0.05 th 1 LTC: Handle backward penalty exactly the same as if it was an isolated pawn. Test for "simplification". See discussion on https://github.com/Rocky640/Stockfish/commit/5d1071da0e57290baa67b9bfe90c0543682ded1b
15-03-19 Roc BackwardLever diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11583 W: 2160 L: 2235 D: 7188
sprt @ 15+0.05 th 1 See what happen if we allow backward to be lever, just as we do for isolated and unsupported.
15-03-19 vin doubled_passer_2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 8311 W: 1539 L: 1623 D: 5149
sprt @ 15+0.05 th 1 Since increasing the penalty caused a fast STC fail, try adjusting the other way.
15-03-19 lan hanging_up diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 21816 W: 4174 L: 4255 D: 13387
sprt @ 15+0.05 th 1 Increase the mg value of hanging
15-03-19 sni sorting_moves9 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 6167 W: 1174 L: 1264 D: 3729
sprt @ 15+0.05 th 1 Try urgent captures first
15-03-19 vin doubled_passer_2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 11035 W: 2045 L: 2121 D: 6869
sprt @ 15+0.05 th 1 Return to the original penalty that passed STC and failed yellow at LTC, but only apply it in full to wing pawns.
15-03-19 sni sorting_moves10 diff
LLR: -2.98 (-2.94,2.94) [-1.50,4.50]
Total: 10354 W: 1936 L: 2015 D: 6403
sprt @ 15+0.05 th 1 Try easy captures first (captures of "hanging" pieces)
15-03-20 bin time_tuning diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11928 W: 2268 L: 2342 D: 7318
sprt @ 15+0.05 th 1 Use stable tuning result
15-03-20 sni easy_captures diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 30037 W: 5738 L: 5764 D: 18535
sprt @ 15+0.05 th 1 Try easy captures first. Take 2, sorting bonus = +10
15-03-20 sni easy_captures diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4402 W: 805 L: 899 D: 2698
sprt @ 15+0.05 th 1 Try easy captures first. Take 3, sorting bonus = +800
15-03-20 vin doubled_passer_measure_ diff
ELO: -2.01 +-2.2 (95%) LOS: 3.6%
Total: 40000 W: 8090 L: 8321 D: 23589
40000 @ 9+0.05 th 1 Measure elo of the original (STC-passing) patch at vSTC to see if there is enough leverage for SPSA.
15-03-20 vin doubled_passer_measure_ diff
ELO: -0.05 +-2.2 (95%) LOS: 48.1%
Total: 40000 W: 8211 L: 8217 D: 23572
40000 @ 9+0.05 th 1 Measure Elo of the very original STC passing patch - this is more CPU-efficient so may have more gain and be better for tuning, although the tuned values need to then be put into the "correct" but slightly slower implementation. Priority -1.
15-03-20 gli master diff
ELO: 15.92 +-1.8 (95%) LOS: 100.0%
Total: 40000 W: 6628 L: 4796 D: 28576
40000 @ 60+0.05 th 1 Regression test against SF6, 8moves_v3
15-03-20 Voy cmhPP diff
LLR: 2.94 (-2.94,2.94) [-1.50,4.50]
Total: 31303 W: 6216 L: 6025 D: 19062
sprt @ 15+0.05 th 1 Add penalty for prior move that allow cutoff.
15-03-20 lan sitting_on_border diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 13266 W: 2560 L: 2630 D: 8076
sprt @ 15+0.05 th 1 Penalty for sitting on a corner with no place to go. Suggested by 'dragon'
15-03-20 lan bishop_check diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 12172 W: 2277 L: 2392 D: 7503
sprt @ 15+0.05 th 1 Modified bishop safe check bonus. Suggested by 'dragon'
15-03-20 Voy cmhPP2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2118 W: 362 L: 463 D: 1293
sprt @ 15+0.05 th 1 Add allow cutoff penalty to both cmh and history tables.
15-03-21 lbr cmh diff
LLR: -2.95 (-2.94,2.94) [-0.50,4.50]
Total: 8597 W: 1592 L: 1688 D: 5317
sprt @ 15+0.05 th 1 linear bonus for cmh. suggested on the forum