Stockfish Testing Queue

Finished - 33100 tests

16-08-31 luc pinned_blocking_pawn diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 13656 W: 2514 L: 2581 D: 8561
sprt @ 10+0.1 th 1 Bonus when an opponent's pinned piece blocks a pawn: take 1
16-08-26 Fis lazyDelta diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 66531 W: 10915 L: 10776 D: 44840
sprt @ 5+0.05 th 7 Lower delta for depths that have already been completed by other threads. Insipred by http://tests.stockfishchess.org/tests/view/57bfe06d0ebc59030fbe4935 bench: 7038730
16-08-31 luc open_ahead diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 37349 W: 6887 L: 6915 D: 23547
sprt @ 10+0.1 th 1 File is semiopen/open for our rook even when there are friendly pawns behind...
16-09-01 sni complexity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11942 W: 2151 L: 2225 D: 7566
sprt @ 10+0.1 th 1 Take 4
16-08-31 sg pawn_history diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 131631 W: 24384 L: 23942 D: 83305
sprt @ 10+0.1 th 1 Add new history table index by pawn structure for quiet move ordering
16-08-29 mco thread_cm diff
ELO: -2.46 +-3.2 (95%) LOS: 6.5%
Total: 9896 W: 1037 L: 1107 D: 7752
10000 @ 60+0.6 th 7 Test per-thread counterMoveHistory, people reported that in case of big NUMA machines, like TCEC superfinal one, this could lead to big nps improvement. We need at least 7 threads and at least 60 secs/game to have a sensible measurement of the possible ELO loss. Set with high priority because TCEC superfinal is near and we need many tests to validate this solution.
16-08-30 mco thread_cm diff
ELO: 3.49 +-4.3 (95%) LOS: 94.6%
Total: 4880 W: 490 L: 441 D: 3949
5000 @ 60+0.6 th 22 Test per-thread counterMoveHistory: Take 2 with 22 threads
16-09-01 Voy pt2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14464 W: 2626 L: 2689 D: 9149
sprt @ 10+0.1 th 1 Take 2
16-09-01 SC highMobility diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 6264 W: 1134 L: 1270 D: 3860
sprt @ 10+0.1 th 1 Should mobility be increased?
16-09-01 Ser threefold_precalculatio diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 46936 W: 8838 L: 8759 D: 29339
sprt @ 10+0.1 th 1 Threefold repetition detection using precalculation. (I tested similar idea long ago, but want to retest it on the current codebase.)
16-09-01 luc pinned_blocking_pawn2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 35921 W: 6590 L: 6561 D: 22770
sprt @ 10+0.1 th 1 Bonus when an opponent's pinned piece blocks a pawn: take 3 (meanwhile, I'll put take 2 to pri -1)
16-09-01 Voy pt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10355 W: 1871 L: 1952 D: 6532
sprt @ 10+0.1 th 1 Make sure that probcut threshold > 0
16-09-01 luc pinned_blocking_pawn diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 48371 W: 8950 L: 8867 D: 30554
sprt @ 10+0.1 th 1 Bonus when an opponent's pinned piece blocks a pawn: take 2
16-08-31 lbr exclusion diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 31023 W: 4187 L: 4080 D: 22756
sprt @ 60+0.6 th 1 LTC: exclusion key should be different depending on which move is excluded
16-09-01 SC simpleImbalance diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 24482 W: 3352 L: 3237 D: 17893
sprt @ 60+0.6 th 1 Bugfix. LTC.
16-09-01 sni complexity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22430 W: 4076 L: 4105 D: 14249
sprt @ 10+0.1 th 1 Change to queenSide * kingSide instead of min(queenSide,kingSide)
16-08-31 Voy cmhBonus diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 26511 W: 4860 L: 4871 D: 16780
sprt @ 10+0.1 th 1 Extra bonuses for cmhs if a non-pv quiet move refutes a pv move.
16-09-01 sg material_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13262 W: 2438 L: 2506 D: 8318
sprt @ 10+0.1 th 1 try the opposite and half table size (8 key bits)
16-08-31 SC mersenne diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 23040 W: 4022 L: 4217 D: 14801
sprt @ 10+0.1 th 1 Now c++11 standard provides us with a reproducible mandated implementation of the mersenne twister. So maybe we can retire our home-made xorshift*. Ah, and this fixes a bug in pick_best, too.
16-08-31 sg material_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17774 W: 3210 L: 3259 D: 11305
sprt @ 10+0.1 th 1 Double up table by using 10 key bits instead of 9
16-08-31 sni complexity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11266 W: 2003 L: 2080 D: 7183
sprt @ 10+0.1 th 1 Use the minimum between number of queenside pawns and number of kingside pawns as a measure of complexity in initiative(). Take 2.
16-08-31 Elb king_centr2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 63957 W: 11813 L: 11749 D: 40395
sprt @ 10+0.1 th 1 Take 3
16-08-31 aji threat_reduction4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 50933 W: 9475 L: 9380 D: 32078
sprt @ 10+0.1 th 1 One more try: STC
16-08-31 SC simpleImbalance diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 43596 W: 8105 L: 8023 D: 27468
sprt @ 10+0.1 th 1 Bugfix.
16-08-31 X frac_reduce diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18608 W: 2438 L: 2495 D: 13675
sprt @ 60+0.6 th 1 Rebase Stefano80's fractional reductions patch. LTC because it didn't show any gain at STC (the only other source of sub-ply reduction is the history score, which I assume is too small at STC). Fixed compilation.
16-08-31 sni complexity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20185 W: 3637 L: 3676 D: 12872
sprt @ 10+0.1 th 1 Use the minimum between number of queenside pawns and number of kingside pawns as a measure of complexity in initiative()
16-08-30 lbr exclusion diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 59814 W: 11136 L: 11083 D: 37595
sprt @ 10+0.1 th 1 exclusion key should be different depending on which move is excluded
16-08-30 sni open_files diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 49112 W: 9239 L: 8900 D: 30973
sprt @ 10+0.1 th 1 Tweak space formula for open files
16-08-31 sg rook_on_file diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 27545 W: 5083 L: 5145 D: 17317
sprt @ 10+0.1 th 1 More endgame bonus for rook on open file (Inspired by the last TCEC game Stockfish - Komodo). (I resubmit the test because of wrong bounds)
16-08-31 jos probcut_us diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 12983 W: 2367 L: 2479 D: 8137
sprt @ 10+0.1 th 1 Take 3.
16-08-31 pb0 penalize2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12165 W: 2208 L: 2281 D: 7676
sprt @ 10+0.1 th 1 Extra penalty for a top quiet (not-TT) move in previous ply when it gets refuted. Can't explain why, but this patch scores so well on my machine, that I like to test it here.
16-08-31 aji threat_reduction3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9646 W: 1707 L: 1791 D: 6148
sprt @ 10+0.1 th 1 Take 3 : STC
16-08-30 aji threat_reduction diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 19389 W: 2507 L: 2562 D: 14320
sprt @ 60+0.6 th 1 Less reduction when in threat: LTC
16-08-30 Voy signAgreement2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 24908 W: 4521 L: 4540 D: 15847
sprt @ 10+0.1 th 1 Take 2...
16-08-30 aji threat_reduction diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 20263 W: 3791 L: 3579 D: 12893
sprt @ 10+0.1 th 1 Less reduction when in threat: STC
16-08-30 Elb king_centr2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 118522 W: 22098 L: 21843 D: 74581
sprt @ 10+0.1 th 1 Take 2: adjust the values in the psqt table
16-08-30 Voy penalizeQuiet' diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10775 W: 1914 L: 1993 D: 6868
sprt @ 10+0.1 th 1 Take 3
16-08-30 Voy penalizeQuiet2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10555 W: 1860 L: 1940 D: 6755
sprt @ 10+0.1 th 1 Take 2
16-08-30 aji threat_extension1 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8527 W: 1524 L: 1613 D: 5390
sprt @ 10+0.1 th 1 Don't go into qsearch when we are threatened by pawn: STC
16-08-30 pec center diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 25258 W: 4101 L: 4125 D: 17032
sprt @ 5+0.05 th 7 Somehow I am getting good result locally with it
16-08-30 Voy penalizeQuiet' diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 39472 W: 7324 L: 7279 D: 24869
sprt @ 10+0.1 th 1 If the 1st move is a quiet and the 2nd a capture cutoff...penalize that quiet move.
16-08-30 jos probcut_us diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 66574 W: 12275 L: 12202 D: 42097
sprt @ 10+0.1 th 1 Raise rbeta for us. (I hope sprt[0, 4] is still ok? Inspired by http://tests.stockfishchess.org/tests/view/57c0b0280ebc59030fbe4989)
16-08-30 SC simpleImbalance diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 191 W: 2 L: 177 D: 12
sprt @ 10+0.1 th 1 Actually, the linear contributions in the material imbalance should be already contained in the values. Is it possible to retire them? Push after rebase.
16-08-30 mbo stats_bonus diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7095 W: 1087 L: 1182 D: 4826
sprt @ 10+0.1 th 3 Reduce bonus for stats when using multiple threads.
16-08-30 hxi king1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7542 W: 1371 L: 1464 D: 4707
sprt @ 10+0.1 th 1 add attackUnits for other checks
16-08-30 jos probcut_us diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 7782 W: 1352 L: 1482 D: 4948
sprt @ 10+0.1 th 1 Raise even further. Take 2.
16-08-30 sni queen_and_knight_forks diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 26407 W: 4838 L: 4850 D: 16719
sprt @ 10+0.1 th 1 Take 2: different values for knight and queen forks
16-08-30 sni knight_fork2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22392 W: 4113 L: 4142 D: 14137
sprt @ 10+0.1 th 1 bonus = S(24,24)
16-08-30 sni knight_fork2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 29330 W: 5430 L: 5429 D: 18471
sprt @ 10+0.1 th 1 Take 4, fork bonus = S(30,30)
16-08-30 pb0 penalty diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25008 W: 4586 L: 4604 D: 15818
sprt @ 10+0.1 th 1 Half penalty also for 2nd refuted quiet move in previous ply