Stockfish Testing Queue

Finished - 30509 tests

16-07-08 Voy lmrSEE diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 6635 W: 1228 L: 1325 D: 4082
sprt @ 10+0.1 th 1 Take 2.
16-07-08 sni drawish5 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17313 W: 3294 L: 3344 D: 10675
sprt @ 10+0.1 th 1 Start from ElbertoOne best version and try to continue it. The main difference is that we don't use the drawish scale factor when either side has hanging pieces.
16-07-07 Roc SimplifyKS2 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 12371 W: 2522 L: 2385 D: 7464
sprt @ 10+0.1 th 1 Same as SimplifyKS1 which passed STC, but now rebased on snicolet Tropism3.
16-06-26 pec mult_stop diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 176721 W: 23187 L: 22497 D: 131037
sprt @ 40+0.4 th 3 LTC Tuned values
16-07-07 Fis centerMobility diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5294 W: 957 L: 1060 D: 3277
sprt @ 10+0.1 th 1 Take2 eg only.
16-07-07 Voy lmrSEE diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4288 W: 795 L: 903 D: 2590
sprt @ 10+0.1 th 1 Make sure the SEE value of not moving the piece is less than moving the piece. i.e. choosing the lesser evil.
16-07-07 sni tropism3 diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 8881 W: 1408 L: 1255 D: 6218
sprt @ 60+0.6 th 1 LTC: Take 6bis: same bench as take 6, but implements a shift trick to count both normal attacks and double attacks with a single popcount, hoping that this will help the 32 bits machines in fishtest.
16-07-07 Voy shallowPruning' diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9795 W: 1819 L: 1902 D: 6074
sprt @ 10+0.1 th 1 Use shallow pruning to assist with LMR.
16-07-07 SC branchingFactor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9573 W: 1777 L: 1861 D: 5935
sprt @ 10+0.1 th 1 Another attempt, suggested by Ivan.
16-07-07 Elb end_draw4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 80268 W: 11705 L: 11539 D: 57024
sprt @ 10+0.1 th 1 Another try at this: only look at almost symmetrical pawn chains with 1 pawn difference.
16-07-07 Elb end_draw5 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 16645 W: 2393 L: 2454 D: 11798
sprt @ 10+0.1 th 1 Last try in this series: almost symmetrical pawn chains in end games
16-07-07 pec dedicated diff
ELO: -5.74 +-4.4 (95%) LOS: 0.5%
Total: 8721 W: 1493 L: 1637 D: 5591
10000 @ 3+0.03 th 7 Use last thread for some no skip widening
16-07-07 sni tropism3 diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 48373 W: 9832 L: 9481 D: 29060
sprt @ 10+0.1 th 1 Take 6bis: same bench as take 6, but implements a shift trick to count both normal attacks and double attacks with a single popcount, hoping that this will help the 32 bits machines in fishtest.
16-07-07 Roc SImplifyKS1 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 32133 W: 6325 L: 6224 D: 19584
sprt @ 10+0.1 th 1 See if we can use attackedBy2 to simplify some king safety calculations, Take 1
16-07-06 Fis centerMobility diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 21739 W: 4011 L: 4043 D: 13685
sprt @ 10+0.1 th 1 Bonus for mobility in the center.
16-07-06 sni tropism3 diff
LLR: -0.51 (-2.94,2.94) [0.00,5.00]
Total: 18504 W: 3675 L: 3613 D: 11216
sprt @ 10+0.1 th 1 King tropism using double attacks. Adjusting weights, take 7
16-07-06 Voy sqc diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 23652 W: 4458 L: 4480 D: 14714
sprt @ 10+0.1 th 1 Score Quiet Checks.
16-07-04 Elb end_draw2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 119379 W: 17661 L: 17351 D: 84367
sprt @ 10+0.1 th 1 Take 2: cover all cases where we have symmetrical pawn chains. Note: take 1 (end_draw) is put on the back burner for now (max. 128K games).
16-07-06 pec tunesf diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 67493 W: 12627 L: 12549 D: 42317
sprt @ 10+0.1 th 1 Final tuned values. ( set by uci options )
16-07-06 sni tropism3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16110 W: 3201 L: 3255 D: 9654
sprt @ 10+0.1 th 1 King tropism using double attacks. Adjusting weights, take 6
16-07-05 pec high_density diff
ELO: -1.84 +-4.1 (95%) LOS: 18.8%
Total: 10000 W: 1773 L: 1826 D: 6401
10000 @ 3+0.03 th 7 High density skip patterns seems to perform at least on par with half density, but quality of moves from individual threads should be higher. Try to by several high density thread and compare with pb's result
16-07-06 Voy quietStats diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7061 W: 1330 L: 1426 D: 4305
sprt @ 10+0.1 th 1 Penalize previous quiet move if it was captured by a killer move.
16-07-05 sni tropism3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 57088 W: 11441 L: 11307 D: 34340
sprt @ 10+0.1 th 1 King tropism using double attacks, take 4
16-07-06 Elb end_draw3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 46536 W: 6790 L: 6744 D: 33002
sprt @ 10+0.1 th 1 Take 3: exclude opposite bishops endgames
16-07-05 Voy gqp diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 29163 W: 5446 L: 5502 D: 18215
sprt @ 10+0.1 th 1 Tweak good capture threshold.
16-07-06 Mat timesimplification diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 16221 W: 2941 L: 3127 D: 10153
sprt @ 10+0.1 th 1 Remove a loop from time management, and use slightly different logic.
16-07-06 Voy csv diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6396 W: 1146 L: 1244 D: 4006
sprt @ 10+0.1 th 1 Spin an idea I had awhile back that sg helped code. Use SEE value to sort bad captures...this time don't sort when depth < 3.
16-07-06 Fis highRankMobility diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6745 W: 1241 L: 1338 D: 4166
sprt @ 10+0.1 th 1 Bonus for mobility in high ranks.
16-07-05 Fis splitMobility diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 4285 W: 792 L: 900 D: 2593
sprt @ 10+0.1 th 1 Split mobility area into low and high value squares roughly based on PSQT table.
16-07-05 Voy Evasion' diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 23273 W: 4360 L: 4384 D: 14529
sprt @ 10+0.1 th 1 Score non-capture negative SEE a higher negative value, than ones with a capture negative SEE.
16-07-05 sni tropism3 diff
LLR: -3.13 (-2.94,2.94) [0.00,5.00]
Total: 21453 W: 4160 L: 4198 D: 13095
sprt @ 10+0.1 th 1 King tropism using double attacks, take 5
16-07-05 Voy shellSort diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16943 W: 3090 L: 3143 D: 10710
sprt @ 10+0.1 th 1 One more try... Use gaps 23, 8, 1.
16-07-02 Fis splitMobilityTune diff
24337/25000 iterations
49568/50000 games played
50000 @ 20+0.2 th 1 Fixed parameter names: Split the mobility board into two regions roughly similar to what PSQT considers good or bad. Tune. Low throughput.
16-07-05 Fis firstQuiet diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7465 W: 1356 L: 1450 D: 4659
sprt @ 10+0.1 th 1 Take 2 more efficient version.
16-07-05 SC ParentNodePruning diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 8655 W: 1567 L: 1741 D: 5347
sprt @ 10+0.1 th 1 A less aggressive version.
16-07-05 SC branchingFactor diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 1755 W: 258 L: 376 D: 1121
sprt @ 10+0.1 th 1 Just use the branching factor to set the initial size of the aspiration window.
16-06-30 Elb end_draw diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 132923 W: 13458 L: 13227 D: 106238
sprt @ 60+0.6 th 1 LTC: My take on drawish endgames with equal material and compact pawn chains.
16-07-05 sni tropism3 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 43039 W: 8479 L: 8411 D: 26149
sprt @ 10+0.1 th 1 King tropism using double attacks, take 3.
16-07-02 Voy trueCM diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20120 W: 2749 L: 2799 D: 14572
sprt @ 60+0.6 th 1 I believe this idea will do much better at LTC...since there may be a higher chance of selecting a "false" cm. Low throughput.
16-07-02 SC PNP_tune diff
19820/20000 iterations
40000/40000 games played
40000 @ 10+0.1 th 1 Short tuning session for the my unifying attempt for razoring and futility pruning.
16-07-04 sni komodo4 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14104 W: 2574 L: 2639 D: 8891
sprt @ 10+0.1 th 1 Sprt test for values which gave +2.2 Elo after 20.000 games (only change the strategy when we are winning). Submitted with priority -2.
16-07-03 sni hanging diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 52215 W: 9840 L: 9737 D: 32638
sprt @ 10+0.1 th 1 More or less factor 1.5 from master, instead of factor 2.
16-07-04 jos early_pruning diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15363 W: 2826 L: 2885 D: 9652
sprt @ 10+0.1 th 1 Skip early pruning also for the first iterations. This helps when tuning at short time controls.
16-07-04 Fis firstQuiet diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 9548 W: 1767 L: 1852 D: 5929
sprt @ 10+0.1 th 1 First quiet idea by @VoyagerOne https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/WL-MmOCec5Y
16-07-04 sni komodo4 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8902 W: 1589 L: 1676 D: 5637
sprt @ 10+0.1 th 1 Sprt test for values which gave +2.2 Elo after 20.000 games. Submitted with priority -2.
16-07-04 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16801 W: 3136 L: 3189 D: 10476
sprt @ 10+0.1 th 1 Recognize pieces which defend attacked squares in the king ring. Try smaller weight. Take 4
16-07-04 Voy Evasion diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20082 W: 3677 L: 3716 D: 12689
sprt @ 10+0.1 th 1 Restructure Evasion MovePick...most notably use SEE only on captures.
16-07-04 sni komodo4 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5907 W: 1031 L: 1131 D: 3745
sprt @ 10+0.1 th 1 Sprt test. Tuned values after 99.300 games.
16-07-04 Voy shellSort diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 17946 W: 3366 L: 3413 D: 11167
sprt @ 10+0.1 th 1 Take 2.
16-07-01 sni tune_komodo' diff
48877/50000 iterations
100000/100000 games played
100000 @ 5+0.05 th 1 Tune second model with different strategies for winning/losing. Tuning at 5+0.05.