Stockfish Testing Queue

Finished - 40825 tests

15-08-23 jki conte diff
ELO: -19.49 +-3.6 (95%) LOS: 0.0%
Total: 20000 W: 5000 L: 6121 D: 8879
20000 @ 15+0.05 th 1 A quick test for contempt implementation. Contempt = 50cp.
15-08-24 jos unstoppable diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13628 W: 2561 L: 2627 D: 8440
sprt @ 15+0.05 th 1 Tuned rank bonuses for unstoppable pawns (allowing a knight for both sides).
15-08-23 lbr probcut diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 32505 W: 6102 L: 6086 D: 20317
sprt @ 15+0.05 th 1 take 4
15-08-24 Mys UnP1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5800 W: 1065 L: 1166 D: 3569
sprt @ 15+0.05 th 1 By file???
15-08-23 lbr probcut^^ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 38833 W: 7279 L: 7235 D: 24319
sprt @ 15+0.05 th 1 more conservative probcut: take 1
15-08-22 jos smp6 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 27097 W: 4534 L: 4549 D: 18014
sprt @ 15+0.05 th 3 Increase minimum split depth by one, if the search is going to be reduced. (Lower throughput, as we currently have many smp-patches running.)
15-08-23 jki histgrav diff
LLR: 2.94 (-2.94,2.94) [0.00,5.00]
Total: 20087 W: 3221 L: 3024 D: 13842
sprt @ 60+0.05 th 1 LTC: History gravity
15-08-23 lbr probcut diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13287 W: 2045 L: 2116 D: 9126
sprt @ 60+0.05 th 1 take 3
15-08-23 Mys UnP diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16778 W: 3143 L: 3196 D: 10439
sprt @ 15+0.05 th 1 Rank base unsupported pawn penalty.
15-08-23 jos smp6 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9776 W: 1559 L: 1643 D: 6574
sprt @ 15+0.05 th 3 Add full reduction. Idea by Tom Vijlbrief. (Take 2) If you think to better test with 7 threads, please tell me.
15-08-23 Voy SelfAdjBonuses diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 7287 W: 1336 L: 1431 D: 4520
sprt @ 15+0.05 th 1 Take 2: This version seems very strong. Changed priority to -1 for original version.
15-08-23 jki histgrav diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 14416 W: 2787 L: 2598 D: 9031
sprt @ 15+0.05 th 1 History gravity
15-08-23 lbr probcut diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 12122 W: 2352 L: 2173 D: 7597
sprt @ 15+0.05 th 1 take 3
15-08-23 Fis timeContempt diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 7299 W: 1374 L: 1469 D: 4456
sprt @ 15+0.05 th 1 See commit notes.
15-08-22 aji max_slaves diff
ELO: 1.74 +-3.6 (95%) LOS: 82.5%
Total: 10000 W: 1457 L: 1407 D: 7136
10000 @ 15+0.05 th 9 Increase the max slaves per splitpoint if depth is more: STC (If you think there is a good reason this is not a good idea, pls comment and I will review/rewrite/stop test) : STC
15-08-23 lbr probcut^ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6137 W: 1085 L: 1184 D: 3868
sprt @ 15+0.05 th 1 take 2
15-08-22 lbr razor diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 38737 W: 7249 L: 7476 D: 24012
sprt @ 15+0.05 th 1 simplify razoring: do not special case depth 1.
15-08-22 Voy AgeStats diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 28426 W: 5318 L: 5320 D: 17788
sprt @ 15+0.05 th 1 Fixed a major flaw when decaying the values. (Thanks to Stefan G for pointing that out)
15-08-22 Roc 2Bs diff
ELO: -2.50 +-3.5 (95%) LOS: 8.4%
Total: 15000 W: 3010 L: 3118 D: 8872
30000 @ 9+0.05 th 1 A BIshop pair experiment
15-08-21 Fis fastWeight diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 75815 W: 14528 L: 14319 D: 46968
sprt @ 15+0.05 th 1 See commit notes.
15-08-13 lbr noCentralBind diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 218170 W: 33620 L: 34101 D: 150449
sprt @ 60+0.05 th 1 Compensate in PSQT
15-08-22 Voy AgeStats diff
LLR: -0.13 (-2.94,2.94) [0.00,5.00]
Total: 627 W: 90 L: 93 D: 444
sprt @ 60+0.05 th 1 I would like to see how well this scale...as local testing seems good. Thoughtput: 250. Will stop early if no progress is being made...
15-08-22 tvi prune diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18341 W: 3496 L: 3541 D: 11304
sprt @ 5+0.05 th 3 Test futility condition TC 5/0.05
15-08-22 Dis castling_bonus diff
LLR: -1.18 (-2.94,2.94) [0.00,5.00]
Total: 17760 W: 3306 L: 3279 D: 11175
sprt @ 15+0.05 th 1 Adding a penality for mid game when the king has not castled yet.
15-08-22 Mys Q1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 10760 W: 1988 L: 2108 D: 6664
sprt @ 15+0.05 th 1 Forward Queen. Take 2
15-08-19 tvi prune diff
LLR: -3.58 (-2.94,2.94) [0.00,5.00]
Total: 49041 W: 8267 L: 8219 D: 32555
sprt @ 15+0.05 th 3 Alternate way of preventing short living pruned splits
15-08-21 aji cutoff_abort diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11827 W: 1933 L: 2009 D: 7885
sprt @ 15+0.05 th 3 Abort the search immediately if a cutoff occurs at a parent splitnode : STC
15-08-21 mco ext_score diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12297 W: 2324 L: 2396 D: 7577
sprt @ 15+0.05 th 1 Experimental patch to do fast weight in the correct way (and even faster)
15-08-17 lbr test diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 90101 W: 14147 L: 14025 D: 61929
sprt @ 60+0.05 th 1 LTC: Less futility + more SE (10442788): - stop futility pruning one depth lower - start singular extension one depth lower - compensate by retuning futility_margin() => Hopefully this can be both an elo gain and an improvement in tactics.
15-08-21 Dis pawn_bishop_penalities_ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15245 W: 2869 L: 2928 D: 9448
sprt @ 15+0.05 th 1 Take 2, thanks again to Rocky for helping with values.
15-08-20 Mys Q diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 30176 W: 5600 L: 5595 D: 18981
sprt @ 15+0.05 th 1 I feel having the queen forward in endgames will provide more options around 3-folds, picking up pawns, checks and attacking lines.
15-08-20 Dis pawn_bishop_penalities_ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6450 W: 1210 L: 1308 D: 3932
sprt @ 15+0.05 th 1 New way of calculating scores for pawns on the same color as a bishop, different score when they are in front or behind the bishop. (Big fix thanks to Rocky)
15-08-20 Voy hStatsTweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15392 W: 2396 L: 2459 D: 10537
sprt @ 60+0.05 th 1 LTC: Take 2.
15-08-20 Voy hStatsTweak diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 27512 W: 5326 L: 5078 D: 17108
sprt @ 15+0.05 th 1 Take 2.
15-08-19 Fis b760017c5ca4dca7cb37fe7 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 22719 W: 4225 L: 4303 D: 14191
sprt @ 15+0.05 th 1 Take 2. Tweak TT replacement policy to use depth divided by age. 2MB STC
15-08-19 Voy AgeStats diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 38612 W: 7232 L: 7190 D: 24190
sprt @ 15+0.05 th 1 Take 2.
15-08-18 Voy hStatsTweak diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 64972 W: 12178 L: 12021 D: 40773
sprt @ 15+0.05 th 1 History stats tweak
15-08-18 lbr epcastle diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 83750 W: 15722 L: 15711 D: 52317
sprt @ 15+0.05 th 1 Prune castling. SEE now handles castling correctly, so there is no reason to single out these moves.
15-08-19 tvi prune diff
LLR: -1.13 (-2.94,2.94) [0.00,5.00]
Total: 2534 W: 430 L: 467 D: 1637
sprt @ 15+0.05 th 3 Less move count pruning for multiple threads
15-08-18 Voy AgeStats diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29232 W: 5495 L: 5493 D: 18244
sprt @ 15+0.05 th 1 Use a count based method for aging history stats.
15-08-19 Roc AdvSquareControl diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 7069 W: 1298 L: 1394 D: 4377
sprt @ 15+0.05 th 1 Endgame bonus for advanced square control.
15-08-18 tvi prune diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 18141 W: 3416 L: 3463 D: 11262
sprt @ 15+0.05 th 1 Take 2
15-08-18 tvi prune diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3210 W: 540 L: 652 D: 2018
sprt @ 15+0.05 th 1 Less move count based pruning in endgame
15-08-14 Roc SupportedPieces diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 123909 W: 23814 L: 23589 D: 76506
sprt @ 15+0.05 th 1 Last try: a fix on take # 3
15-08-17 lbr tune diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 102829 W: 19517 L: 19313 D: 63999
sprt @ 15+0.05 th 1 tuned RazorMargin and FutilityMargin
15-08-16 Roc SupportedKnight diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 108703 W: 20677 L: 20494 D: 67532
sprt @ 15+0.05 th 1 This time, using exactly the same simple bonus and code as the SupportedPiece test, but only for Knight. Last try on this.
15-08-17 lbr tune diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 47192 W: 8823 L: 8816 D: 29553
sprt @ 15+0.05 th 1 tuned locally (4+0.04 30k iter) RazorMargin and FutilityMargin. as the fishtest values are quite different from the values obtained locally, let's try the locally tuned values too. If both fail, it means I am probably "tuning noise", as these parameters are not sensitive enough.
15-08-18 Voy lmrTweak diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22972 W: 4328 L: 4353 D: 14291
sprt @ 15+0.05 th 1 LMR Tweak
15-08-18 jhe lmr_npv diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 5193 W: 844 L: 947 D: 3402
sprt @ 30+0.05 th 1 Create a floor for reductions on non-pv nodes.
15-08-16 lbr tune diff
34455/30000 iterations
69906/70000 games played
70000 @ 15+0.05 th 1 TUNE(RazorMargin, FutilityMargin)