Stockfish Testing Queue

Finished - 22407 tests

10-02-14 rs iid_tweak1 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 28894 W: 5356 L: 5386 D: 18152
sprt @ 15+0.05 th 1 Do IID search also if ttMove is from qsearch
10-02-14 jo razor_margin diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 6975 W: 1554 L: 1644 D: 3777
sprt @ 15+0.05 th 1 Try SPSA values for razor_margin. Take 3.
10-02-14 rs probcut diff
LLR: -1.03 (-2.94,2.94) [-1.50,4.50]
Total: 7507 W: 1382 L: 1399 D: 4726
sprt @ 15+0.05 th 1 Attempt to improve move ordering for probcut search
10-02-14 ur null_modify diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 21558 W: 3953 L: 4002 D: 13603
sprt @ 15+0.05 th 1 second try of null move pruning change we do not use null move pruning when eval<beta and I want also not to use null move pruning when result of qsearch is smaller than beta by the same logic
10-02-14 ho master diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 46790 W: 8541 L: 8525 D: 29724
sprt @ 15+0.05 th 1 Avoid calls to pos.legal() in most cases in search()
10-02-14 rs null_intermediate diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 4119 W: 760 L: 844 D: 2515
sprt @ 60+0.05 th 1 Search at intermediate depth if remaining depth is high. Take 1
10-02-14 ur null_modify diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 9439 W: 1437 L: 1496 D: 6506
sprt @ 60+0.05 th 1 trying to do always qsearch in null move pruning with the idea that starting with qsearch before search help to get better order of moves.
10-02-14 rs null_intermediate diff
LLR: -1.44 (-2.94,2.94) [-1.50,4.50]
Total: 6522 W: 1191 L: 1225 D: 4106
sprt @ 15+0.05 th 1 Search at intermediate depth if remaining depth is high. Take 2
10-02-14 rs null_intermediate diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 19321 W: 3643 L: 3488 D: 12190
sprt @ 15+0.05 th 1 Search at intermediate depth if remaining depth is high. Take 1
10-02-14 ur null_modify diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 12528 W: 2343 L: 2206 D: 7979
sprt @ 15+0.05 th 1 trying to do always qsearch in null move pruning with the idea that starting with qsearch before search help to get better order of moves.
08-02-14 vd measure_razor_margins2 diff
ELO: 0.76 +-1.9 (95%) LOS: 78.3%
Total: 40000 W: 6249 L: 6162 D: 27589
40000 @ 60+0.05 th 1 This passed STC with a clear elo gain. ELO: 2.61 +-2.1 (95%) LOS: 99.3%. Yet it failed SPRT(0,6) at LTC rather quickly. Is this the mytical non-scalable patch? I would like to make a more precise measurement of the actual elo at LTC. Low priority.
09-02-14 rs null_tweak diff
LLR: -2.95 (-2.94,2.94) [0.00,6.00]
Total: 12722 W: 2380 L: 2418 D: 7924
sprt @ 60+0.05 th 1 Final take 2. Direct LTC because TC dependent. 2moves_v1book. Low prio.
09-02-14 jo c_checks_stm diff
ELO: 0.43 +-2.4 (95%) LOS: 63.8%
Total: 40000 W: 9985 L: 9935 D: 20080
40000 @ 5+0.05 th 1 Measure stm bonus for contact checks
10-02-14 in pv_instability diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2132 W: 337 L: 436 D: 1359
sprt @ 15+0.05 th 1 Take 5: Decay PV faster when depth is greater. Little bit twisted, but works at short TC.
09-02-14 in pv_instability diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1500 W: 228 L: 329 D: 943
sprt @ 15+0.05 th 1 Take 4: Decay PV faster when depth is greater
09-02-14 Fi checkextless_pvinstabil diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 22359 W: 4104 L: 4151 D: 14104
sprt @ 15+0.05 th 1 Combo of check_ext_less and pv_instability both of which passed STC and were positive on LTC.
09-02-14 in razor_margin diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7089 W: 1286 L: 1373 D: 4430
sprt @ 15+0.05 th 1 Try to make razor_margin more quadratic with respect to <depth> and see if it scales better. Also, use vd's tweaked pre-condition.
09-02-14 in razor_margin diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 1342 W: 209 L: 311 D: 822
sprt @ 15+0.05 th 1 Extreme Try: Try to make razor_margin more exponential with respect to <depth> and see if it scales better. Also, use vd's tweaked pre-condition.
09-02-14 in razor_margin diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1800 W: 322 L: 425 D: 1053
sprt @ 5+0.05 th 1 Short TC: Extreme Try: Try to make razor_margin more exponential with respect to <depth> and see if it scales better. Also, use vd's tweaked pre-condition.
09-02-14 rs null_tweak diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 3129 W: 517 L: 613 D: 1999
sprt @ 15+0.05 th 1 attempt to smuggle Uri's max depth condition for nullmove search. Take 1
08-02-14 hw see_king_capture diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 81337 W: 15060 L: 14745 D: 51532
sprt @ 15+0.05 th 1 Another attempted see optimization.
09-02-14 in razor_margin diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7107 W: 1444 L: 1532 D: 4131
sprt @ 5+0.05 th 1 Shorter TC: Try to make razor_margin more quadratic with respect to <depth> and see if it scales better. Also, use vd's tweaked pre-condition.
09-02-14 lb attack_unit diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 20345 W: 4581 L: 4630 D: 11134
sprt @ 15+0.05 th 1 do not cap attackUnit
09-02-14 sg lmr_exclude diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14438 W: 2598 L: 2666 D: 9174
sprt @ 15+0.05 th 1 LMR: less reduction (half ply) for followup moves
09-02-14 sg lmr_exclude diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 16541 W: 2993 L: 3055 D: 10493
sprt @ 15+0.05 th 1 LMR: less reduction (half ply) for double checks
08-02-14 ur fix_null_bug diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 35355 W: 6400 L: 6414 D: 22541
sprt @ 40/15 th 1 This change cause stockfish not to be blind to zugzwangs.(I guess that at 40/15 it can get the needed depth in endgames to detect some zugzwangs and play better moves.
08-02-14 mc 0a1092f64eb9eb232094 diff
LLR: 2.96 (-2.94,2.94) [-4.00,0.00]
Total: 38903 W: 6482 L: 6452 D: 25969
sprt @ 15+0.05 th 3 Verify racy simplification (SMP with 3 threads)
08-02-14 in razor_margin diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 8317 W: 1493 L: 1576 D: 5248
sprt @ 15+0.05 th 1 Try tuning with CLOP @ 10+0.05 to see if it scales better.
07-02-14 lu tricky_delta diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 42950 W: 7976 L: 7969 D: 27005
sprt @ 15+0.05 th 1 reset delta based on depth and BestMoveChanges - take 2
07-02-14 jo razor_margin^ diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 29474 W: 5444 L: 5472 D: 18558
sprt @ 15+0.05 th 1 Reducing at remaining depth == 1 is much less risky than at depth == 3. So I start with a much smaller margin. Take 1.
07-02-14 in from_null diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14452 W: 2590 L: 2658 D: 9204
sprt @ 15+0.05 th 1 Reintroduce fromNull (from Mindbreaker's potential regressions list -- edce2a8 -- -29elo?). Instead of testing old patch vs parent master, I decided to try and reintroduce it into latest master.
07-02-14 rs null_verification diff
LLR: -1.53 (-2.94,2.94) [-4.00,0.00]
Total: 53660 W: 9794 L: 10062 D: 33804
sprt @ 15+0.05 th 1 solves 8/8/8/2p5/1pp5/brpp4/1pprp2P/qnkbK3 w - - 0 1 at depth 98. Take 2. Simplification test this time.
08-02-14 rs null_verification diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7968 W: 1436 L: 1520 D: 5012
sprt @ 15+0.05 th 1 solves 8/8/8/2p5/1pp5/brpp4/1pprp2P/qnkbK3 w - - 0 1 at depth 96. Double verification search. Take 3 (final)
07-02-14 vd measure_razor_margins3 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 57336 W: 10565 L: 10521 D: 36250
sprt @ 15+0.05 th 1 Razor margins are now roughly 1/3 of precondition margins.
07-02-14 lu 9e327f28c21320da9ea849a diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 32167 W: 5927 L: 5948 D: 20292
sprt @ 15+0.05 th 1 reset delta based on depth and BestMoveChanges - take 1
08-02-14 pe tm diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 21844 W: 3960 L: 4009 D: 13875
sprt @ 15+0.05 th 1 Roughly proportional increase in thinking time with faster decay of pv instability time
08-02-14 vd measure_razor_margins3 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 17772 W: 3307 L: 3366 D: 11099
sprt @ 15+0.05 th 1 Razor margins are now roughly 2/3 of precondition margins. Last attempt in this series.
07-02-14 ur lessnull2 diff
ELO: -3.06 +-2.2 (95%) LOS: 0.3%
Total: 35777 W: 6412 L: 6727 D: 22638
40000 @ 15+0.05 th 1 I multiply by ONE_PLY that I forgot in the previous version (the previous version was not my intention but it has smaller bench so maybe it is better so it may be interesting to test both the previous version and also my intention for comparison).
07-02-14 in razor_futility_margin diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 15686 W: 2878 L: 2942 D: 9866
sprt @ 15+0.05 th 1 Tune razor_margin and futility_margin. This time I try using CLOP instead of SPSA.
07-02-14 rs null_verification diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 8115 W: 1448 L: 1532 D: 5135
sprt @ 15+0.05 th 1 make null move verification search more accurate so that 8/8/8/2p5/1pp5/brpp4/1pprp2P/qnkbK3 w - - 0 1 can be solved fast
07-02-14 gl measure_razor_margins2 diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 21267 W: 3272 L: 3278 D: 14717
sprt @ 60+0.05 th 1 LTC SPRT for VD: This time I have set the razor margins to half of the precondition margins.
07-02-14 ur lessnull2 diff
ELO: -15.38 +-3.8 (95%) LOS: 0.0%
Total: 12000 W: 1948 L: 2479 D: 7573
40000 @ 15+0.05 th 1 limit using null move pruning(simpler version then previous version because I use number of nodes to decide about big depth not to prune only at the beginning of the iteration.
07-02-14 vd measure_razor_margins2 diff
ELO: 2.61 +-2.1 (95%) LOS: 99.3%
Total: 40000 W: 7565 L: 7264 D: 25171
40000 @ 15+0.05 th 1 This time I have set the razor margins to half of the precondition margins.
06-02-14 do c4533e0d94 diff
ELO: 1.98 +-2.9 (95%) LOS: 90.8%
Total: 20000 W: 3754 L: 3640 D: 12606
20000 @ 15+0.05 th 1 Verify possible regression from Mindbreaker's list
07-02-14 jo razor_margin diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 15034 W: 2732 L: 2798 D: 9504
sprt @ 15+0.05 th 1 New razor_margin, take 2.
07-02-14 vd measure_razor_margins diff
ELO: -4.38 +-2.1 (95%) LOS: 0.0%
Total: 38796 W: 6933 L: 7422 D: 24441
40000 @ 15+0.05 th 1 I suspect that the use of razor_margins(depth) both as a precondition for razoring and as actual margins is wrong. To measure this I have set the margins to zero but kept the precondition.
07-02-14 Th checkinfo diff
LLR: -0.07 (-2.94,2.94) [-1.50,4.50]
Total: 238 W: 39 L: 41 D: 158
sprt @ 15+0.05 th 1 Pass CheckInfo by reference rather than const reference for speed gain
06-02-14 sg lmr_exclude diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7274 W: 1283 L: 1369 D: 4622
sprt @ 15+0.05 th 1 LMR: exclude passed pawn pushes
06-02-14 sg lmr_exclude diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7564 W: 1337 L: 1422 D: 4805
sprt @ 15+0.05 th 1 LMR: less reductions for passed pawn pushes
06-02-14 hx scaling diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 60928 W: 11256 L: 10994 D: 38678
sprt @ 15+0.05 th 1 scaling with 2 and 3 pawns - take 3