Stockfish Testing Queue

Finished - 20083 tests

31-08-13 pe mc_pruning diff
LLR: -2.95 (-2.94,2.94)
Total: 20715 W: 3635 L: 3635 D: 13445
sprt @ 60+0.05 th 1 Fix all out of place points in move count pruning curves. Previous simulations showed that curves are close to optimum, but fixing just 2 out of place points at low depth gave around 2 true elo across long and short TC. Here I fix all out of place points. They are at higher depth so I test only at 60+0.05, as at 15+0.05 I consider it essentially tested.
31-08-13 ur more_time_for_big_chang diff
LLR: -2.95 (-2.94,2.94)
Total: 3658 W: 731 L: 829 D: 2098
sprt @ 15+0.05 th 1 using more time when the score is changed significantly
31-08-13 ur union_changes2 diff
LLR: 2.96 (-2.94,2.94)
Total: 18936 W: 3723 L: 3566 D: 11647
sprt @ 30+0.05 th 1 this is a union of 2 changes and I choose 30+0.05 for first stage because of the claim that one of the changes is good only in long time control and I think that in this case it is better to use 30+0.05 for stage 1 and not to skip it because we need stage 1 for some confidence that the change works.
31-08-13 tv King diff
LLR: -2.96 (-2.94,2.94)
Total: 45951 W: 9651 L: 9628 D: 26672
sprt @ 15+0.05 th 1 Tune rank 3 shield
30-08-13 do master diff
ELO: 105.04 +-3.4 (95%) LOS: 100.0%
Total: 20000 W: 8044 L: 2175 D: 9781
20000 @ 60+0.05 th 1 Let's measure total progress made in fishtest as it is probably already over +100 ELO now!
30-08-13 gl pawn_tune diff
ELO: -4.45 +-3.1 (95%) LOS: 0.2%
Total: 20000 W: 4017 L: 4273 D: 11710
20000 @ 15+0.05 th 1 Double pawn chain values
31-08-13 rt SwordFish_A2 diff
ELO: -34.40 +-3.4 (95%) LOS: 0.0%
Total: 20000 W: 3876 L: 5850 D: 10274
20000 @ 15+0.05 th 1 Give SwordFish a second chance at playing against Base. I hope that the difference is < 15 ELO points, but it may be more or less than my guess.
30-08-13 mc skill_level^^^^ diff
ELO: 115.61 +-4.9 (95%) LOS: 100.0%
Total: 19380 W: 11864 L: 5643 D: 1873
20000 @ 15+0.05 th 1 Skill Level 6 vs 5
30-08-13 mc skill_level diff
ELO: 44.89 +-4.4 (95%) LOS: 100.0%
Total: 19785 W: 9432 L: 6890 D: 3463
20000 @ 15+0.05 th 1 Skill Level 10 vs 9
31-08-13 jk timesqrt diff
LLR: -2.96 (-2.94,2.94)
Total: 2562 W: 505 L: 607 D: 1450
sprt @ 15+0.05 th 1 sqrt(bestMoveChanges) + faster aging
31-08-13 jk timesqrt diff
LLR: -2.96 (-2.94,2.94)
Total: 3734 W: 746 L: 844 D: 2144
sprt @ 15+0.05 th 1 sqrt(bestMoveChanges)
30-08-13 gl blocked diff
LLR: -2.97 (-2.94,2.94)
Total: 7942 W: 1618 L: 1704 D: 4620
sprt @ 15+0.05 th 1 Knights get small bonus for opponent blocked pawn
30-08-13 mc skill_level^^^ diff
ELO: 106.96 +-4.8 (95%) LOS: 100.0%
Total: 19000 W: 11228 L: 5557 D: 2215
20000 @ 15+0.05 th 1 Skill Level 7 vs 6
30-08-13 gl blocked diff
LLR: -2.94 (-2.94,2.94)
Total: 2547 W: 537 L: 640 D: 1370
sprt @ 15+0.05 th 1 Knights get big bonus for opponent blocked pawns
30-08-13 gl blocked diff
LLR: -2.97 (-2.94,2.94)
Total: 4979 W: 1032 L: 1127 D: 2820
sprt @ 15+0.05 th 1 Knights get bonus for opponent blocked pawns
31-08-13 mc search diff
LLR: -2.95 (-2.94,2.94)
Total: 5920 W: 1187 L: 1278 D: 3455
sprt @ 15+0.05 th 1 Alternative unstoppable eval: take 1 (I go crazy thinking that we have more than 160 lines of evaluation code for a +3 ELO)
31-08-13 ee MoreLMR diff
LLR: -2.96 (-2.94,2.94)
Total: 14503 W: 3000 L: 3067 D: 8436
sprt @ 15+0.05 th 1 Gary's patch was so successful, that maybe we could try some more moves. I remember though that the killer moves were not always excluded from LMR, but that may have been an unofficial Stockfish.
30-08-13 mc skill_level^^ diff
ELO: 92.03 +-4.6 (95%) LOS: 100.0%
Total: 20000 W: 11218 L: 6041 D: 2741
20000 @ 15+0.05 th 1 Skill Level 8 vs 7
30-08-13 gl blocked diff
LLR: -2.97 (-2.94,2.94)
Total: 10108 W: 2142 L: 2222 D: 5744
sprt @ 15+0.05 th 1 Penalty for isolated and blocked pawns
30-08-13 mc skill_level^ diff
ELO: 69.77 +-4.5 (95%) LOS: 100.0%
Total: 20000 W: 10401 L: 6438 D: 3161
20000 @ 15+0.05 th 1 Skill Level 9 vs 8
31-08-13 do old_patches diff
LLR: -2.96 (-2.94,2.94)
Total: 4961 W: 1028 L: 1123 D: 2810
sprt @ 15+0.05 th 1 Old so-called "fromNull" patch (eventually reverted). Check with SPRT.
31-08-13 ee AlwaysDancing_part1 diff
LLR: -2.95 (-2.94,2.94)
Total: 2813 W: 587 L: 689 D: 1537
sprt @ 15+0.05 th 1 Another try at the futility margin in case of negative SEE There should be a working version, somewhere. But there could be dependancy on timecontrol also.
30-08-13 do old_patches diff
LLR: -2.95 (-2.94,2.94)
Total: 8669 W: 1757 L: 1840 D: 5072
sprt @ 15+0.05 th 1 Old patch "Check for easy move just once" (eventually reverted). Check with SPRT.
30-08-13 jk ext diff
LLR: -2.96 (-2.94,2.94)
Total: 6520 W: 1339 L: 1429 D: 3752
sprt @ 15+0.05 th 1 Extend all dangerous moves
30-08-13 rt SwordFish_A1 diff
LLR: -2.69 (-2.94,2.94)
Total: 28883 W: 6061 L: 6077 D: 16745
sprt @ 15+0.05 th 1 Push up mobility to compensate for the removal of PSQT (SwordFish vs SwordFish Test). If this test fails, it will give me a valid range for this branch's tuned mobility value. If it passes, the range needs to be extended further.
30-08-13 mc skill_level diff
ELO: 105.14 +-4.8 (95%) LOS: 100.0%
Total: 19982 W: 12119 L: 6250 D: 1613
20000 @ 10+0.05 th 1 Skill Level 5 vs 4
30-08-13 gl easy1 diff
LLR: -2.96 (-2.94,2.94)
Total: 14811 W: 2575 L: 2605 D: 9631
sprt @ 60+0.05 th 1 Uri test: LongTC - changing easy move code in order not to have easy moves in case that the program changed its mind in one of the last iterations.
25-08-13 jk onProbCut diff
ELO: -12.91 +-2.8 (95%) LOS: 0.0%
Total: 19575 W: 2956 L: 3683 D: 12936
20000 @ 150+0.05 th 1 Test if ProbCut is harmful with long time control
30-08-13 jk nolmrext diff
LLR: -2.97 (-2.94,2.94)
Total: 6950 W: 1433 L: 1522 D: 3995
sprt @ 15+0.05 th 1 Disable LMR for extended moves
30-08-13 mc skill_level^^ diff
ELO: 113.97 +-4.9 (95%) LOS: 100.0%
Total: 20000 W: 12656 L: 6321 D: 1023
20000 @ 10+0.05 th 1 Skill Level 3 vs 2
30-08-13 mc skill_level^^^ diff
ELO: 141.48 +-5.1 (95%) LOS: 100.0%
Total: 20000 W: 13525 L: 5803 D: 672
20000 @ 10+0.05 th 1 Skill Level 2 vs 1
30-08-13 mc skill_level^ diff
ELO: 122.97 +-4.9 (95%) LOS: 100.0%
Total: 20000 W: 12759 L: 5962 D: 1279
20000 @ 10+0.05 th 1 Skill Level 4 vs 3
29-08-13 mc master diff
ELO: 16.18 +-2.9 (95%) LOS: 100.0%
Total: 20000 W: 4152 L: 3221 D: 12627
20000 @ 60+0.05 th 1 Our first regression of new dev branch. Until Gary's LMR of dangerous moves.
22-08-13 tk Multiple-Not-Improving diff
LLR: -2.95 (-2.94,2.94)
Total: 20959 W: 3920 L: 3915 D: 13124
sprt @ 60+0.05 th 1 Long TC: Try again with /10
28-08-13 ur easy1 diff
LLR: 2.95 (-2.94,2.94)
Total: 61420 W: 12779 L: 12498 D: 36143
sprt @ 15+0.05 th 1 changing easy move code in order not to have easy moves in case that the program changed its mind in one of the last iterations.
30-08-13 rt SwordFish diff
ELO: -47.48 +-3.5 (95%) LOS: 0.0%
Total: 20000 W: 3975 L: 6691 D: 9334
20000 @ 15+0.05 th 1 SwordFish is a little pet project I have decided to partake in where I am attempting to eradicate unneeded PSQT entries from Stockfish. Please allow this test and all of the games run through to give me an estimate of how far behind (as it will likely be) SwordFish is from Stockfish. Anecdotal tests give me reason to be optimistic, but it is horrible for opening analysis so it needs work regardless.
27-08-13 pe mc_pruning diff
LLR: -2.95 (-2.94,2.94)
Total: 99615 W: 20647 L: 20473 D: 58495
sprt @ 15+0.05 th 1 Couple more shots at long end of passed stage I patch: very small bump down
30-08-13 sg eval_interpolation diff
LLR: -2.95 (-2.94,2.94)
Total: 6030 W: 1239 L: 1330 D: 3461
sprt @ 15+0.05 th 1 use quadratic polynomial with a=0.001
30-08-13 sg eval_interpolation diff
LLR: -2.96 (-2.94,2.94)
Total: 19673 W: 4136 L: 4188 D: 11349
sprt @ 15+0.05 th 1 use quadratic polynomial with a=-0.001
30-08-13 sg eval_interpolation diff
LLR: -2.96 (-2.94,2.94)
Total: 7228 W: 1489 L: 1577 D: 4162
sprt @ 15+0.05 th 1 use tangens with b=0.01
30-08-13 sg eval_interpolation diff
LLR: -2.96 (-2.94,2.94)
Total: 11066 W: 2345 L: 2422 D: 6299
sprt @ 15+0.05 th 1 use arctangens with b=0.01
30-08-13 sg eval_interpolation diff
LLR: -2.95 (-2.94,2.94)
Total: 2480 W: 495 L: 597 D: 1388
sprt @ 15+0.05 th 1 I think the linear eval interpolation function is not optimal. First try, use sinus instead
30-08-13 ee AlwaysDancing_part1 diff
LLR: -2.94 (-2.94,2.94)
Total: 3835 W: 798 L: 896 D: 2141
sprt @ 15+0.05 th 1 I would describe this as the central part of AlwaysDancing. There are still many changes but they do go together in my opinion. The futility margin is from Rainbow Serpent. I have only 100 games so I'm setting priority -2. I do try to get some more games in my own testing.
23-08-13 pe mc_pruning diff
LLR: -2.96 (-2.94,2.94)
Total: 84900 W: 15822 L: 15486 D: 53592
sprt @ 60+0.05 th 1 LTC: prune more nonimproved nodes at low depth. It passed stage I but am not sure if this is enough to pass stage II
29-08-13 gl less_futility diff
LLR: -2.95 (-2.94,2.94)
Total: 2766 W: 550 L: 651 D: 1565
sprt @ 15+0.05 th 1 Higher starting futility
29-08-13 gl less_futility diff
LLR: -2.96 (-2.94,2.94)
Total: 5466 W: 1112 L: 1205 D: 3149
sprt @ 15+0.05 th 1 Linear scaling of futility margin
29-08-13 ur time6 diff
LLR: -2.95 (-2.94,2.94)
Total: 37475 W: 7851 L: 7852 D: 21772
sprt @ 15+0.05 th 1 test using more time at time trouble relative to previous version(the difference at depth n is proportional to 0.8^n)
29-08-13 mc check_prune3 diff
LLR: 2.97 (-2.94,2.94)
Total: 16441 W: 3102 L: 2912 D: 10427
sprt @ 60+0.05 th 1 LTC: Allow LMR on dangerous moves
29-08-13 ur limit_unstability_time diff
LLR: -2.97 (-2.94,2.94)
Total: 7625 W: 1576 L: 1663 D: 4386
sprt @ 15+0.05 th 1 testing limiting the target time for unstable moves.
29-08-13 jh queen_tune diff
LLR: -2.95 (-2.94,2.94)
Total: 7687 W: 1543 L: 1629 D: 4515
sprt @ 15+0.05 th 1 Tune queen base values.