Stockfish Testing Queue

Finished - 22438 tests

28-09-14 sg cpp diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 45708 W: 9502 L: 9265 D: 26941
sprt @ 15+0.05 th 1 Try a more conservative approach. Add only 25% to middle and endgame bonus.
28-09-14 jo always_ks diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 9844 W: 1957 L: 2037 D: 5850
sprt @ 15+0.05 th 1 Always compute king-safety.
28-09-14 mb connected_pp2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11588 W: 2313 L: 2388 D: 6887
sprt @ 15+0.05 th 1 Use only a file based bonus for connected passers. My last try on this.
28-09-14 mb connected_pp diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 28142 W: 5783 L: 5811 D: 16548
sprt @ 15+0.05 th 1 Another attempt at connected passed pawns. Rescheduled. Deleted previous scheduled run by accident. :P
28-09-14 lb ttDepth diff
LLR: -2.96 (-2.94,2.94) [-1.00,4.00]
Total: 14260 W: 2780 L: 2869 D: 8611
sprt @ 15+0.05 th 1 Correct ttDepth: 3rd phase for recaptures. Can be viewed as a bugfix, but test instead for a small elo gain (as it adds a bit of probably useless code).
28-09-14 My pssd_pwns1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 1762 W: 337 L: 497 D: 928
sprt @ 15+0.05 th 1 Like outposts try to tune passed pawns for mid/end game. Low Pri take 2.
28-09-14 My pssd_pwns diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 958 W: 160 L: 325 D: 473
sprt @ 15+0.05 th 1 Like outposts try to tune passed pawns for mid/end game. Low Pri
28-09-14 sg cpp diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1712 W: 329 L: 434 D: 949
sprt @ 15+0.05 th 1 Doubled bonus for connected passers in endgame + little middle game bonus
28-09-14 sg cpp diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2306 W: 454 L: 557 D: 1295
sprt @ 15+0.05 th 1 Doubled bonus for connected passers in endgame
28-09-14 mb conn_pp_closeby diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 6023 W: 1184 L: 1274 D: 3565
sprt @ 15+0.05 th 1 Add bonus to all connected passers only if they are close to each other.
28-09-14 My con_pwns1 diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 16484 W: 3339 L: 3438 D: 9707
sprt @ 15+0.05 th 1 Like outposts try to tune connected pawns for mid/end game. Low Pri take 2.
28-09-14 lb recapture^^ diff
ELO: -1.50 +-2.7 (95%) LOS: 13.5%
Total: 29267 W: 6458 L: 6584 D: 16225
30000 @ 7.5+0.025 th 1 DEPTH_QS_RECAPTURE = -6
28-09-14 mb wing_pp_closeby diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 7130 W: 1427 L: 1515 D: 4188
sprt @ 15+0.05 th 1 Add bonus to connected outside passers only if they are close to each other.
28-09-14 lb recapture^ diff
ELO: -0.60 +-2.6 (95%) LOS: 32.6%
Total: 30000 W: 6658 L: 6710 D: 16632
30000 @ 7.5+0.025 th 1 DEPTH_QS_RECAPTURE = -4
28-09-14 lb recapture diff
ELO: -3.09 +-2.7 (95%) LOS: 1.2%
Total: 28346 W: 6189 L: 6441 D: 15716
30000 @ 7.5+0.025 th 1 DEPTH_QS_RECAPTURE = -7
28-09-14 My con_pwns diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 13779 W: 2848 L: 2958 D: 7973
sprt @ 15+0.05 th 1 Like outposts try to tune connected pawns for mid/end game. Low Pri
23-09-14 ee null_verification_12ply diff
ELO: -1.92 +-2.3 (95%) LOS: 5.2%
Total: 30000 W: 5161 L: 5327 D: 19512
30000 @ 60+0.05 th 1 Low priority. Base signature was wrong. Because the signatures actually are quite different, maybe worth some testing.
27-09-14 sg pruning diff
LLR: -3.67 (-2.94,2.94) [-1.50,4.50]
Total: 42668 W: 8679 L: 8693 D: 25296
sprt @ 15+0.05 th 1 exclude counter moves from pruning
27-09-14 gl wing_pp diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 10153 W: 2084 L: 2163 D: 5906
sprt @ 15+0.05 th 1 Connected pawns, lower weight
26-09-14 ur lesspartial diff
ELO: 3.87 +-3.1 (95%) LOS: 99.3%
Total: 20000 W: 4255 L: 4032 D: 11713
20000 @ 15+0.05 th 1 This change make the depth always even. The target is to measure no partial extensions(to see if there is no significant reduction) if not it may be possible to simplify the code later and get rid of the confusing ONE_PLY=2(if there is a significant reduction that I need to check the source of it because I did some changes)
27-09-14 sg pruning diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 71332 W: 14687 L: 14594 D: 42051
sprt @ 15+0.05 th 1 exclude follow up moves from pruning
27-09-14 gl wing_pp diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2328 W: 460 L: 563 D: 1305
sprt @ 15+0.05 th 1 Connected passed pawns are a dangerous long term threat (higher weights)
27-09-14 gl wing_pp diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 6636 W: 1327 L: 1416 D: 3893
sprt @ 15+0.05 th 1 Connected passed pawns are a dangerous long term threat (noticed from under eval in Gull TCEC game from season 7, round 1)
27-09-14 sg pruning diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1200 W: 202 L: 307 D: 691
sprt @ 15+0.05 th 1 exclude moves with history > 0 from pruning
27-09-14 sg pruning diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 27311 W: 5664 L: 5694 D: 15953
sprt @ 15+0.05 th 1 exclude killers from pruning
26-09-14 ur lesspartial diff
ELO: 0.26 +-2.2 (95%) LOS: 59.1%
Total: 38898 W: 8021 L: 7992 D: 22885
40000 @ 15+0.05 th 1 After removing some partial reduction in LMR I want to test removing it from null move pruning when the final target is to have no partial reductions so we can use ONE_PLY=1 (note that in order to have no partial reduction in LMR I need also to fix the array of reduction not to allow partial reductions)
26-09-14 jo smp1 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 35186 W: 6661 L: 6673 D: 21852
sprt @ 15+0.05 th 3 Another try to widen the search with multiple threads. Low pri. Take 2.
27-09-14 ur lesspartial diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 7647 W: 1356 L: 1214 D: 5077
sprt @ 60+0.05 th 1 Test the same change at long time control to see if it is also improvement at long time control. use (-3,1) because it is a simplification in case that it pass.(considering the good result at short time control with 20,000 games as pass at short time control)
26-09-14 jo smp1 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 32363 W: 6157 L: 6177 D: 20029
sprt @ 15+0.05 th 3 Another try to widen the search with multiple threads. Low pri.
26-09-14 lb qspvs diff
LLR: -2.95 (-2.94,2.94) [-1.00,4.00]
Total: 14566 W: 2913 L: 3001 D: 8652
sprt @ 15+0.05 th 1 PVS in qsearch(), with aggressive re-search (ie. neglect instability)
26-09-14 aj tune_min_time diff
LLR: -3.04 (-2.94,2.94) [0.00,4.00]
Total: 35900 W: 7370 L: 7401 D: 21129
sprt @ 15+0.05 th 1 Try out values obtained from SPSA run for Min Thinking time: STC
26-09-14 lb qspvs diff
LLR: -2.96 (-2.94,2.94) [-1.00,4.00]
Total: 10072 W: 2035 L: 2136 D: 5901
sprt @ 15+0.05 th 1 PVS in qsearch(), if first move gains material
26-09-14 lb qspvs^ diff
LLR: -2.95 (-2.94,2.94) [-1.00,4.00]
Total: 15312 W: 3094 L: 3180 D: 9038
sprt @ 15+0.05 th 1 PVS in qsearch()
25-09-14 ur safety3 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 44949 W: 7846 L: 7852 D: 29251
sprt @ 60+0.05 th 1 It is the same change as ksafety2 but from the latest master and I use (0,4) as people suggested because it is practically similar to a parameter tuning(does not do the code more complex)
24-09-14 ur no_verification2 diff
ELO: -0.62 +-2.0 (95%) LOS: 26.7%
Total: 39199 W: 6314 L: 6384 D: 26501
40000 @ 120+0.05 th 1 trying no verification search(lucas say that abs is not needed but I prefer simply not to do null move pruning in these cases to be sure not to get a wrong score(in practical games the condition about abs(beta) almost never happens and I was not sure if I need beta<VALUE_KNOWN_WIN or beta>-VALUE_KNOWN_WIN to avoid problems and I prefer simply to avoid pruning in these cases.
25-09-14 ur ksafety2 diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 12526 W: 2706 L: 2562 D: 7258
sprt @ 15+0.05 th 1 another try to do king safety only in part of the cases(this time in relatively less cases in order to make the program faster)
25-09-14 ur ksafety2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 25058 W: 5182 L: 5219 D: 14657
sprt @ 15+0.05 th 1 Trying a simplified version that is almost equivalent(same bench but not equivalent based on bench at higher depths) when I use the stage of the game directly(and not material on the board when different materials can have the stage of the game) .
25-09-14 ur maybe_better_verificati diff
ELO: -2.00 +-2.2 (95%) LOS: 3.8%
Total: 40000 W: 8343 L: 8573 D: 23084
40000 @ 15+0.05 th 1 check a different idea for verification search from depth 6 and check first to see if there is no big regression at short time control because I start to do verification search from depth 6
24-09-14 ur ksafety2 diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 15250 W: 3335 L: 3182 D: 8733
sprt @ 15+0.05 th 1 do almost always king safety evaluation(I save the effort if the weight of the king safety evaluation is too small because the position is close to the endgame so calculating it is probably too expensive to get the benefit)
24-09-14 My K_OP2 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4447 W: 907 L: 1003 D: 2537
sprt @ 15+0.05 th 1 Outpost bonus when square not attacked or cannot be attacked by enemy pawn.
24-09-14 My K_OP diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 8286 W: 1664 L: 1748 D: 4874
sprt @ 15+0.05 th 1 Give a higher outpost bonus when the outpost square is not attacked by an enemy pawn.(Deleted previous as hash was uneven)
24-09-14 lb reductions diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 27162 W: 4802 L: 4692 D: 17668
sprt @ 60+0.05 th 1 LTC for Uri: simplification in order to get rid of half ply reductions
21-09-14 ur no_verification diff
ELO: 1.08 +-2.0 (95%) LOS: 85.3%
Total: 40000 W: 7027 L: 6903 D: 26070
40000 @ 60+0.05 th 1 test value of verification search at long time control(I suspect that verification search is counter productive at long time control and I want to test the value of it at 60+0.05(we tested it only a long time ago and it was without 2moves_v1.pgn and it is possible that things are different today and we get clearly significant bad result) I keep verification search when there is a known win but it almost never happen.
24-09-14 ur reductions diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 9829 W: 2142 L: 1998 D: 5689
sprt @ 15+0.05 th 1 simplification in order to get rid of half ply reductions
21-09-14 gl null diff
ELO: -0.14 +-1.8 (95%) LOS: 44.0%
Total: 40000 W: 5557 L: 5573 D: 28870
40000 @ 60+0.05 th 1 Measure at LTC: Limit evaluation based null move extra reduction to three plies
22-09-14 lb cleanup diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 42579 W: 7580 L: 7496 D: 27503
sprt @ 60+0.05 th 1 Cleanup VALUE_KNOWN_WIN conditions
22-09-14 jk ksafety diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 40101 W: 7011 L: 7234 D: 25856
sprt @ 60+0.05 th 1 STC: Always init king safety
22-09-14 lb cleanup diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 14575 W: 3101 L: 2967 D: 8507
sprt @ 15+0.05 th 1 Cleanup VALUE_KNOWN_WIN conditions
21-09-14 ee null_verification_8ply diff
ELO: -6.68 +-2.3 (95%) LOS: 0.0%
Total: 30000 W: 4920 L: 5497 D: 19583
30000 @ 60+0.05 th 1 Measure starting verification search at 8 ply with Joseph's R. LTC because we need to hit sufficient depth even if we start earlier and patch must scale.
21-09-14 jk ksafety diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 86881 W: 17951 L: 17953 D: 50977
sprt @ 15+0.05 th 1 STC: Always init king safety