Stockfish Testing Queue

Finished - 20718 tests

08-03-14 rs lmr_red_inc diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 3216 W: 539 L: 635 D: 2042
sprt @ 15+0.05 th 1 More conservative reduction for early moves. Variety 2
08-03-14 rs lmr_red_inc diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11582 W: 2093 L: 2168 D: 7321
sprt @ 15+0.05 th 1 More conservative reduction for early moves. Variety 1
08-03-14 My Probcut diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 31128 W: 5674 L: 5699 D: 19755
sprt @ 15+0.05 th 1 Cut node attempt
08-03-14 ok cutNode1 diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 4780 W: 669 L: 749 D: 3362
sprt @ 60+0.05 th 1 LTC for jo: The first node after a null move is a cut node. Let's try it.
08-03-14 pe predict_first diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 9976 W: 1816 L: 1895 D: 6265
sprt @ 15+0.05 th 1 soft stop if hard stop due to first root move is expected
08-03-14 mc retire_cutNode diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 26612 W: 4903 L: 4792 D: 16917
sprt @ 15+0.05 th 1 Retest setting cutNode = true in null verification, this time with bug fix SPRT parameters because in null verification we expect a fail high (>99% of cases)
08-03-14 rs ks_check_remove2 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14741 W: 2692 L: 2759 D: 9290
sprt @ 15+0.05 th 1 Remove most of the check tests. Take 2
07-03-14 vd simpl diff
ELO: 0.34 +-2.2 (95%) LOS: 61.9%
Total: 40000 W: 8272 L: 8233 D: 23495
40000 @ 5+0.05 th 1 Start removing preconditions to razoring to see if the elo gain at STC persists. Very fast time control as a preliminary test.
07-03-14 jo cutNode1 diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 45664 W: 8308 L: 8087 D: 29269
sprt @ 15+0.05 th 1 The first node after a null move is a cut node. Let's try it.
06-03-14 sn understand_cutNode diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 34186 W: 6177 L: 6194 D: 21815
sprt @ 15+0.05 th 1 understand_cutNode (take 7 : in ProbCuct eval >= beta + 150). Low priority.
07-03-14 My Se_tweak diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7782 W: 1393 L: 1478 D: 4911
sprt @ 15+0.05 th 1 Tweak depth again with different approach. Included mstembera's tip
07-03-14 My se_tweak diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 12865 W: 2303 L: 2375 D: 8187
sprt @ 15+0.05 th 1 Tweak depth / 3 * 2
06-03-14 sn understand_cutNode diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 19761 W: 3702 L: 3546 D: 12513
sprt @ 15+0.05 th 1 understand_cutNode (take 5 : in ProbCuct eval >= beta + 100, forcing nodeCut = false). Low priority.
06-03-14 sn understand_cutNode diff
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 73684 W: 13680 L: 13384 D: 46620
sprt @ 15+0.05 th 1 understand_cutNode (take 4 : in ProbCuct eval >= beta + 50, forcing nodeCut = false). Low priority.
05-03-14 vd razor_tune_try diff
ELO: 0.66 +-1.7 (95%) LOS: 78.1%
Total: 50000 W: 7578 L: 7483 D: 34939
50000 @ 60+0.05 th 1 CLOP values do not appear to be changing a lot anymore so let's try LTC for this succesful STC test. Fixed number of games since experience in the past seems to indicate razoring may not be scalable. I would like so see some elo estimate at the end instead of just pass/fail.
06-03-14 sn understand_cutNode diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 26725 W: 4879 L: 4915 D: 16931
sprt @ 15+0.05 th 1 understand_cutNode (take 3 : in product eval >= beta + 50). Low priority
05-03-14 sn retire_cutNode diff
LLR: -0.19 (-2.94,2.94) [-1.50,4.50]
Total: 128000 W: 23521 L: 23198 D: 81281
sprt @ 15+0.05 th 1 take 6
06-03-14 sn understand_cutNode diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 22651 W: 4092 L: 4139 D: 14420
sprt @ 15+0.05 th 1 understand_cutNode (take 2 : LMR extension = 3/2). Low priority.
06-03-14 sn understand_cutNode diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 7649 W: 1388 L: 1473 D: 4788
sprt @ 15+0.05 th 1 understand_cutNode (take 1, LMR ext = 3 /4). Low priority
27-02-14 pe 0949f06 diff
ELO: 38.93 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 9407 L: 4944 D: 25649
40000 @ 60+0.05 th 1 Regression test until Fix a warning with Intel compiler right before Dynamic draw. Low priority.
04-03-14 pe contempt_no_phase diff
ELO: 166.57 +-2.6 (95%) LOS: 100.0%
Total: 40000 W: 20617 L: 2786 D: 16597
40000 @ 15+0.05 th 1 Measure elo against stockfish 3 with no contempt. Low priority
05-03-14 pe still_first diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 8696 W: 1544 L: 1626 D: 5526
sprt @ 15+0.05 th 1 Chip away more time from unchanging moves. take 2
05-03-14 sn retire_null_move diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 225 W: 12 L: 118 D: 95
sprt @ 15+0.05 th 1 Even more probabilistic cuts (take 2)
05-03-14 sn retire_null_move diff
LLR: -5.67 (-2.94,2.94) [-1.50,4.50]
Total: 250 W: 1 L: 182 D: 67
sprt @ 15+0.05 th 1 Even more probabilistic cuts
04-03-14 sn retire_cutNode diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 7746 W: 1349 L: 1434 D: 4963
sprt @ 15+0.05 th 1 take 5
04-03-14 sn retire_cutNode diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4533 W: 791 L: 884 D: 2858
sprt @ 15+0.05 th 1 take 4
04-03-14 sn retire_cutNode diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2598 W: 427 L: 525 D: 1646
sprt @ 15+0.05 th 1 take 3
04-03-14 sn retire_cutNode diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 1350 W: 206 L: 308 D: 836
sprt @ 15+0.05 th 1 take 2
04-03-14 vd razor_tune_try_extra_pl diff
ELO: 1.70 +-2.1 (95%) LOS: 94.7%
Total: 40000 W: 7438 L: 7242 D: 25320
40000 @ 15+0.05 th 1 Blue sky: measure the potential of an extra ply of razoring.
04-03-14 pe contempt_no_phase diff
ELO: 173.82 +-2.7 (95%) LOS: 100.0%
Total: 40000 W: 21426 L: 2932 D: 15642
40000 @ 15+0.05 th 1 Measure if contempt will add more elo against much weaker opponent SF3 -- around 150 elo weaker than current version. Against stockfish dd it was 5.6 +- 3.1 at 15+0.05. Low priority
04-03-14 vd razor_tune_try diff
ELO: 1.75 +-2.1 (95%) LOS: 95.2%
Total: 40000 W: 7435 L: 7234 D: 25331
40000 @ 15+0.05 th 1 CLOP seems to show a tendency towards converging. Let's try the new values.
03-03-14 sn stringent_null_move diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 4642 W: 777 L: 869 D: 2996
sprt @ 15+0.05 th 1 The large reduction depth of null move implies a risk. Try to compensate for that risk by moving the bound a bit (take 2).
03-03-14 sn tweak-probcut-precondit diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 10146 W: 1672 L: 1726 D: 6748
sprt @ 60+0.05 th 1 LTC for : Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut (take 4)
04-03-14 sn retire_cutNode diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2584 W: 411 L: 509 D: 1664
sprt @ 15+0.05 th 1 retire cutNode
01-03-14 mc contempt_no_phase diff
ELO: 41.97 +-2.1 (95%) LOS: 100.0%
Total: 40000 W: 9970 L: 5161 D: 24869
40000 @ 60+0.05 th 1 Regression test at LTC of Leonid's contempt: we expect a gain given SF DD is about 40 ELO weaker
03-03-14 tz remove_doubled diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 27115 W: 5023 L: 5057 D: 17035
sprt @ 15+0.05 th 1 remove doubled (simplification) scored +236 -212 =702 in test 20+0.05
03-03-14 sn tweak-probcut-precondit diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 11221 W: 2131 L: 2207 D: 6883
sprt @ 15+0.05 th 1 Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut (take 5)
03-03-14 rs history_bonus diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18139 W: 3276 L: 3334 D: 11529
sprt @ 15+0.05 th 1 incorporate distance from root into bonus calculation for history update
01-03-14 sn tweak-probcut-precondit diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 64789 W: 12263 L: 12197 D: 40329
sprt @ 15+0.05 th 1 Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut.
01-03-14 mc contempt_no_phase diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 43517 W: 7089 L: 7004 D: 29424
sprt @ 60+0.05 th 1 LTC of Leonid contempt idea tested with 'no regression' mode [-3, 1] against original master (without dynamic draw)
03-03-14 sn tweak-probcut-precondit diff
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 34515 W: 6707 L: 6509 D: 21299
sprt @ 15+0.05 th 1 Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut (take 4)
03-03-14 sn tweak-probcut-precondit diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 11813 W: 2190 L: 2264 D: 7359
sprt @ 15+0.05 th 1 Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut (take 2)
02-03-14 sn tweak-probcut-precondit diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 14254 W: 2675 L: 2743 D: 8836
sprt @ 15+0.05 th 1 Try to avoid searching the probcut subtrees when the searches are not likely to produce a cut (take 3)
02-03-14 tz remove_undefendedMinors diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 10809 W: 2060 L: 2137 D: 6612
sprt @ 15+0.05 th 1 remove undefendedMinors (simplification) scored +165 -144 =460 in test 20+0.05
02-03-14 sn stringent_null_move diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 3383 W: 616 L: 713 D: 2054
sprt @ 15+0.05 th 1 The large reduction depth of null move implies a risk. Try to compensate for that risk by moving the bound a bit.
01-03-14 pe still_first diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 37488 W: 5736 L: 5670 D: 26082
sprt @ 60+0.05 th 1 LTC. Chip away more time from unchanging moves
27-02-14 mc master diff
ELO: 39.25 +-2.1 (95%) LOS: 100.0%
Total: 40000 W: 9664 L: 5164 D: 25172
40000 @ 60+0.05 th 1 Standard LTC regression test until Dynamic draw
01-03-14 rs rcc_try diff
LLR: -2.97 (-2.94,2.94) [0.00,6.00]
Total: 3747 W: 564 L: 649 D: 2534
sprt @ 60+0.05 th 1 Attempt for more precise check bonuses. Final take 2. Direct LTC because TC sensitive. Reduced priority.
01-03-14 vd razor_tune_try diff
ELO: -0.19 +-2.1 (95%) LOS: 42.8%
Total: 40000 W: 7341 L: 7363 D: 25296
40000 @ 15+0.05 th 1 Some parameters found by CLOP. CLOP has not converged yet but I would like to see some confirmation that it is at least going in the right direction.
27-02-14 mc master diff
ELO: 37.55 +-2.5 (95%) LOS: 100.0%
Total: 25079 W: 5548 L: 2848 D: 16683
40000 @ 60+0.05 th 3 Standard LTC regression test until Dynamic draw (SMP 3 threads) to verify possible SMP bugs