Stockfish Testing Queue

Finished - 24196 tests

14-10-14 jo noTempo diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 10226 W: 2043 L: 2122 D: 6061
sprt @ 15+0.05 th 1 No tempo bonus in endgame. Take 2.
14-10-14 jo noTempo^^ diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 7377 W: 1474 L: 1561 D: 4342
sprt @ 15+0.05 th 1 No tempo bonus in endgame. Take 1
11-10-14 fw easyMove@zeroCost6 diff
LLR: -0.11 (-2.94,2.94) [0.00,6.00]
Total: 57928 W: 9881 L: 9605 D: 38442
sprt @ 60+0.05 th 1 simiplified version one with hard coded critical_number_of_stable_moves
12-10-14 jo KingDanger diff
ELO: -0.37 +-2.3 (95%) LOS: 37.4%
Total: 40000 W: 8932 L: 8975 D: 22093
40000 @ 7+0.05 th 1 Test new values for KingDanger. Low pri, because most probably only worth 0 - 1.5 elo. Too difficult to measure locally.
14-10-14 lu history_aware_timemanag diff
LLR: -0.01 (-2.94,2.94) [-1.50,4.50]
Total: 8682 W: 1802 L: 1778 D: 5102
sprt @ 15+0.05 th 1 Take into account some stats from previous move when determining current move's importance: take 1
13-10-14 aj multithreats_tuned_B diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 56551 W: 11652 L: 11386 D: 33513
sprt @ 15+0.05 th 1 Use SPSA tuned values. But zero out the negative values : STC
13-10-14 lb space diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 21168 W: 3611 L: 3696 D: 13861
sprt @ 60+0.05 th 1 tuned weights
13-10-14 aj multithreats_tuned_A diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 6968 W: 1349 L: 1437 D: 4182
sprt @ 15+0.05 th 1 SPSA tuned values of multithreats: STC
13-10-14 lb space diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 15535 W: 3316 L: 3096 D: 9123
sprt @ 15+0.05 th 1 tuned weights
11-10-14 ur easy_move6 diff
ELO: -2.18 +-3.1 (95%) LOS: 8.7%
Total: 18497 W: 3602 L: 3718 D: 11177
20000 @ 15+0.05 th 1 Trying to test more aggressive reduction in time in case the pawn difference is at least 1/8 pawns(twice faster) because it seems time management is not sensitive to small changes.
12-10-14 lb space diff
53500/50000 iterations
109216/110000 games played
110000 @ 15+0.05 th 1 tune space related weights
12-10-14 ur easy7_test diff
ELO: -27.55 +-13.6 (95%) LOS: 0.0%
Total: 1049 W: 179 L: 262 D: 608
20000 @ 15+0.05 th 1 Simply play twice faster when you did not change your mind in many iterations and I suspect that this is the reason that the easy move based on previous depth get good results.
12-10-14 do outpost diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 51424 W: 10513 L: 10448 D: 30463
sprt @ 15+0.05 th 1 Outpost simplification with locally tuned parameters
12-10-14 Fi TT_probecount diff
ELO: -1.65 +-3.0 (95%) LOS: 14.4%
Total: 20000 W: 3939 L: 4034 D: 12027
20000 @ 15+0.05 th 1 Prefer TT entries that have been probed more often. Tuning 1
12-10-14 Fi TT_probed diff
ELO: 1.11 +-3.1 (95%) LOS: 76.1%
Total: 20000 W: 4087 L: 4023 D: 11890
20000 @ 15+0.05 th 1 Prefer TT entries that have been probed at least once. Tuning 2. Pri -1
12-10-14 aj spsa_tune_multithreats diff
48647/50000 iterations
100000/100000 games played
100000 @ 15+0.05 th 1 Multiple threats probably need an additional bonus. Difficult to arrive at correct values without tuning.
12-10-14 sn keep_material6 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2586 W: 465 L: 565 D: 1556
sprt @ 15+0.05 th 1 Last try : best version so far in local testing
12-10-14 aj spsa_tune_multithreats diff
131/50000 iterations
260/100000 games played
100000 @ 15+0.05 th 1 Multiple threats probably need an additional bonus. Difficult to arrive at correct values without tuning.
11-10-14 aj multiple_threats diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 9985 W: 1976 L: 2056 D: 5953
sprt @ 15+0.05 th 1 Additional bonus for multiple threats. Use a better set of values: STC
11-10-14 sn king_support diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4018 W: 760 L: 856 D: 2402
sprt @ 15+0.05 th 1 Try a quadratic formula
11-10-14 ur space_change diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 19111 W: 3860 L: 3914 D: 11337
sprt @ 15+0.05 th 1 another try to change space this time I calcualte it in more cases and not only in the early stage of the middle game. It is changing parameters so I plan to use sprt(0,4) at long time control if it pass
09-10-14 ur key8 diff
ELO: -24.76 +-2.9 (95%) LOS: 0.0%
Total: 20000 W: 2898 L: 4321 D: 12781
20000 @ 60+0.05 th 1 If the loss is not bigger relative to shorter time control then for me it is enough to feel safe that in correspondence games there is no problem(assuming you use at least 1024 mbytes with the hardware of today).
11-10-14 Fi TT_probed diff
ELO: -2.19 +-3.0 (95%) LOS: 7.9%
Total: 20000 W: 3923 L: 4049 D: 12028
20000 @ 15+0.05 th 1 Prefer TT entries that have been probed at least once. Tuning 1
11-10-14 ur easy_move6 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 24239 W: 4844 L: 4884 D: 14511
sprt @ 15+0.05 th 1 Testing less aggresive formula(the change is half the change that I tried in previous test that seems not to change much).
10-10-14 ur easy_move6 diff
ELO: 0.05 +-3.0 (95%) LOS: 51.3%
Total: 20000 W: 3992 L: 3989 D: 12019
20000 @ 15+0.05 th 1 Fixing a bug in my previous easy move(I plan to do at most 8 more tries of easy moves before I give up).
11-10-14 sn king_support diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 4208 W: 820 L: 916 D: 2472
sprt @ 15+0.05 th 1 Replace evaluate_unstoppable_pawns() by evaluate_king_support_for_passed_pawns()
11-10-14 sg ks_blocked diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 13268 W: 2638 L: 2708 D: 7922
sprt @ 15+0.05 th 1 Reduce king safety if king side is blocked. Refine blocking conditions and use bigger bonus. Take 3
11-10-14 fw easyMove@zeroCost6 diff
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 18241 W: 3756 L: 3598 D: 10887
sprt @ 15+0.05 th 1 simiplified version one with hard coded critical_number_of_stable_moves
11-10-14 ur space_change diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 62709 W: 12488 L: 12765 D: 37456
sprt @ 15+0.05 th 1 test the value of calculating space only in special conditions and not always(positive score of +2 elo in the fixed game tests after more than 9000 games so I plan to test it with sprt because it is a simplification that remove if conditions).
10-10-14 lb unstoppable diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 57370 W: 11566 L: 11834 D: 33970
sprt @ 15+0.05 th 1 remove unstoppable
10-10-14 sn keep_material6 diff
19735/20000 iterations
40000/40000 games played
40000 @ 15+0.05 th 1 SPSA session for the idea of keeping pawns and exchanging pieces when ahead. Default values seemed reasonable in local testing (540 - 490 - 1515 [0.510] after 2545 games).
10-10-14 ur space_change diff
ELO: 2.05 +-4.5 (95%) LOS: 81.6%
Total: 9315 W: 1888 L: 1833 D: 5594
20000 @ 15+0.05 th 1 test the value of calculating space only in special conditions and not always.
10-10-14 fw easyMove@zeroCost6 diff
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]
Total: 2662 W: 510 L: 611 D: 1541
sprt @ 15+0.05 th 1 similar to version one, but with more aggressive timelimit, for extremely stable preferred move
10-10-14 lb pawnHash diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 35650 W: 7199 L: 7227 D: 21224
sprt @ 15+0.05 th 1 Double PawnHash. non functional change. small speedup in local testing. need to get an idea on longer TC and with representative blend of machines, so test in the framework.
05-10-14 Fi key14 diff
ELO: -0.07 +-2.0 (95%) LOS: 47.3%
Total: 40000 W: 6872 L: 6880 D: 26248
40000 @ 60+0.05 th 1 Verify 14 vs 16 bit ELO loss at LTC 8MB is NOT substantially greater than the -1.84 ELO we measured at STC 2MB. See forum for more details. Low pri.
10-10-14 fw easyMove@zeroCost6 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 1549 W: 260 L: 363 D: 926
sprt @ 15+0.05 th 1 take 3 scale with best move changes. The Idea is to also use use shown stability more aggresively.
09-10-14 ra LMR_A diff
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 71095 W: 12390 L: 12143 D: 46562
sprt @ 60+0.05 th 1 LTC: Reduce brutally moves that enter the square the opponent just left.
10-10-14 Fi key8 diff
ELO: -32.00 +-3.3 (95%) LOS: 0.0%
Total: 20000 W: 3659 L: 5496 D: 10845
20000 @ 7.5+0.05 th 1 8 bit key 7.5+0.05 1MB. Quick cheap test to see if this is even worse than the -28.03elo we saw @ STC 2MB. Pri -3
09-10-14 ur easy_move6 diff
ELO: -13.25 +-3.3 (95%) LOS: 0.0%
Total: 17292 W: 3213 L: 3872 D: 10207
20000 @ 15+0.05 th 1 using difference between latest score of best move and latest score of second best move for time management(earlier I used only latest known difference at the same depth). Relatively aggressive formula of time change(and I am going to test different formula if it fails).
08-10-14 Fi key8 diff
ELO: -23.47 +-2.9 (95%) LOS: 0.0%
Total: 20000 W: 2921 L: 4270 D: 12809
20000 @ 60+0.05 th 1 The results of this test compared to 8 bit STC 2MB will show whether or not TC matters. This is the deciding test. Pri -1
10-10-14 jk sec diff
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]
Total: 14300 W: 2883 L: 2950 D: 8467
sprt @ 15+0.05 th 1 Every second iteration
09-10-14 lb space diff
ELO: -8.83 +-3.1 (95%) LOS: 0.0%
Total: 20235 W: 3900 L: 4414 D: 11921
30000 @ 15+0.05 th 1 measure minor behind pawn. if small, should be simplified away by simply retuning fine-tuning space (and connected).
07-10-14 ur easy_move4 diff
ELO: -1.20 +-3.1 (95%) LOS: 22.2%
Total: 20000 W: 4039 L: 4108 D: 11853
20000 @ 15+0.05 th 1 Same patch as easy_move3 except different formula
08-10-14 fw easyMove@zeroCost6 diff
LLR: -0.74 (-2.94,2.94) [0.00,6.00]
Total: 3841 W: 650 L: 657 D: 2534
sprt @ 60+0.05 th 1 Take 2. Reduce effect when Depth is already very high, such as in endgames.
08-10-14 gl mstembera_ttkey diff
LLR: -6.10 (-2.94,2.94) [-3.00,1.00]
Total: 200399 W: 40008 L: 40711 D: 119680
sprt @ 15+0.05 th 1 Try mstembera 21bit key version at 4mb, no regression
07-10-14 Fi key8 diff
ELO: -28.03 +-3.1 (95%) LOS: 0.0%
Total: 20000 W: 3372 L: 4982 D: 11646
20000 @ 15+0.05 th 1 8 bit key. Trying to determine the minimum key size for TCEC. See forum for more details. Low pri.
08-10-14 Fi key10 diff
ELO: -5.79 +-3.1 (95%) LOS: 0.0%
Total: 20000 W: 3892 L: 4225 D: 11883
20000 @ 15+0.05 th 1 10 bit key. Trying to determine the minimum key size for STC/TCEC. 12 was ok. 8 looks not. See forum for more details. Pri -1.
07-10-14 ur easy_move4 diff
ELO: 0.23 +-3.0 (95%) LOS: 55.8%
Total: 20000 W: 4023 L: 4010 D: 11967
20000 @ 15+0.05 th 1 more risky formula to reduce time for comparison(still not very risky and for evaluation difference of less than 2 pawns at depth 5 I reduce only by factor of less than 2.
08-10-14 sn keep_material6 diff
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]
Total: 14959 W: 3001 L: 3067 D: 8891
sprt @ 15+0.05 th 1 Take 3 : when ahead, keep pawns and exchange pieces
08-10-14 sn doubled_isolated2 diff
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]
Total: 21464 W: 4355 L: 4402 D: 12707
sprt @ 15+0.05 th 1 sprt test with penalty for doubled isolated pawns tuned by SPSA