Stockfish Testing Queue

Finished - 27341 tests

12-02-16 sn centerBind diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25028 W: 4656 L: 4673 D: 15699
sprt @ 10+0.1 th 1 Take 1
12-02-16 ci small_large_stats diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 25406 W: 4867 L: 4630 D: 15909
sprt @ 10+0.1 th 1 The idea is to store two history value per entry. One with a small limit (short-term stats) and another with a large limit (long-term stats).
12-02-16 My qp2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12994 W: 2419 L: 2488 D: 8087
sprt @ 10+0.1 th 1 Likely last take
12-02-16 Vo ev-killers diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18279 W: 3392 L: 3438 D: 11449
sprt @ 10+0.1 th 1 Score both Killers in Evasion move pick.
11-02-16 II SPSA_benchmark_tune diff
91163/100000 iterations
200000/200000 games played
200000 @ 2+0.02 th 1 Take 5 corrected (as explained on the forum) : testing the best obtained parameters with 100K iterations. Information obtained will be used for comparison with values obtained by "noisy DOE", and for SPSA user guidelines. Throughput 250.
11-02-16 II noisy_DOE diff
23947/25000 iterations
50000/50000 games played
50000 @ 2+0.02 th 1 Take 1: c=300, R=0.0011
11-02-16 Vo spt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17413 W: 3197 L: 3247 D: 10969
sprt @ 10+0.1 th 1 Shallow Pruning Tweak
11-02-16 Vo fp-simple diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 13366 W: 2446 L: 2628 D: 8292
sprt @ 10+0.1 th 1 Simplify move count pruning. (See if we can eventually get rid of 'improving')
11-02-16 pb variax60 diff
ELO: 2.02 +-4.0 (95%) LOS: 84.0%
Total: 10000 W: 1733 L: 1675 D: 6592
10000 @ 4+0.04 th 7 TRY 4, (fixed boolean initialization) From the variax results so far (very high residuals through the series) it is not clear if the concept of introducing a second thread beside mainthread not skipping depths (variax2) is a good idea. Since the difference might be small I think the unique way to figure it out is to compare them directly with smaller map & number of threads. This is a test not against master!
11-02-16 II SPSA_benchmark_tune diff
752/100000 iterations
1510/200000 games played
200000 @ 2+0.02 th 1 Take 5 : testing the best obtained parameters with 100K iterations. Information obtained will be used for comparison with values obtained by "noisy DOE", and for SPSA user guidelines. Throughput 250.
11-02-16 II SPSA_benchmark_tune diff
24543/25000 iterations
50000/50000 games played
50000 @ 2+0.02 th 1 Take 4:switching back to alpha = 0.602.
11-02-16 My qp2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 45855 W: 8637 L: 8562 D: 28656
sprt @ 10+0.1 th 1 Take 3
11-02-16 Ro ColourControl diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 19469 W: 3679 L: 3720 D: 12070
sprt @ 10+0.1 th 1 Take 2. Smaller penalty.
11-02-16 My qt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10253 W: 1935 L: 2016 D: 6302
sprt @ 10+0.1 th 1 Try also small danger increase where queen present
11-02-16 pb variax60 diff
ELO: -1.00 +-22.4 (95%) LOS: 46.5%
Total: 346 W: 64 L: 65 D: 217
10000 @ 4+0.04 th 7 TRY 3.
11-02-16 pb variax60 diff
ELO: 49.26 +-46.7 (95%) LOS: 98.2%
Total: 71 W: 17 L: 7 D: 47
10000 @ 4+0.04 th 7 Retry. From the variax results so far (very high residuals through the series) it is not clear if the concept of introducing a second thread beside mainthread not skipping depths (variax2) is a good idea. Since the difference might be small I think the unique way to figure it out is to compare them directly with smaller map & number of threads. This is a test not against master!
11-02-16 pb variax60 diff
ELO: -6.56 +-32.8 (95%) LOS: 34.7%
Total: 106 W: 12 L: 14 D: 80
10000 @ 4+0.04 th 7 From the variax results so far (very high residuals through the series), it is not clear if the concept of introducing a second thread beside mainthread not skipping depths (variax2) brought a benefit or not. Since the difference might be small I think the unique way is to compare them directly with smaller map & number of threads. This is a test not against master!
11-02-16 Vo qCnt-lmr diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5848 W: 1011 L: 1111 D: 3726
sprt @ 10+0.1 th 1 Don't LMR the first quiet move.
11-02-16 Ro ColourControl diff
LLR: -3.02 (-2.94,2.94) [0.00,5.00]
Total: 5752 W: 1022 L: 1126 D: 3604
sprt @ 10+0.1 th 1 Penalty when giving more squares of one colour in our space.
10-02-16 II SPSA_benchmark_tune diff
22471/25000 iterations
50000/50000 games played
50000 @ 2+0.02 th 1 Take 3: the last try to change alpha if there will be no progress.
10-02-16 pb variax2 diff
ELO: 0.80 +-4.0 (95%) LOS: 65.1%
Total: 10000 W: 1761 L: 1738 D: 6501
10000 @ 3+0.03 th 11 Take 2: Trying if this more flexible approach with 11 threads
10-02-16 II SPSA_benchmark_tune diff
22611/25000 iterations
50000/50000 games played
50000 @ 2+0.02 th 1 Take 2: as explained on the forum.
10-02-16 Vo AP-LMR diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 11660 W: 2105 L: 2180 D: 7375
sprt @ 10+0.1 th 1 Take 3...
10-02-16 pb variax2 diff
ELO: 2.85 +-3.8 (95%) LOS: 92.7%
Total: 10000 W: 1641 L: 1559 D: 6800
10000 @ 3+0.03 th 23 Take 2: Trying if this more flexible approach regarding number of threads is even stronger
10-02-16 II SPSA_benchmark_tune diff
23690/25000 iterations
50000/50000 games played
50000 @ 2+0.02 th 1 Testing values obtained by DoE and recalculation.
10-02-16 Vo lmr-root diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11837 W: 2192 L: 2266 D: 7379
sprt @ 10+0.1 th 1 Don't lmr root moves if it has a positive history value.
10-02-16 Vo AP-LMR diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9912 W: 1773 L: 1856 D: 6283
sprt @ 10+0.1 th 1 Take 2
10-02-16 My qd diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9405 W: 1706 L: 1791 D: 5908
sprt @ 10+0.1 th 1 Take 2
10-02-16 My qd diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 26785 W: 4933 L: 4943 D: 16909
sprt @ 10+0.1 th 1 Increased King danger according to enemy Queen proximity
10-02-16 sg rook_on_pawn diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 6379 W: 1163 L: 1262 D: 3954
sprt @ 10+0.1 th 1 Give additional bonus for weak pawns. Take 2
10-02-16 My sm diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 22870 W: 4441 L: 4465 D: 13964
sprt @ 10+0.1 th 1 Add Queen to space weight
10-02-16 sg rook_on_pawn diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8247 W: 1531 L: 1621 D: 5095
sprt @ 10+0.1 th 1 Give RookOnPawn bonus only for weak pawns. Compensate by raising bonus.
10-02-16 Vo ptp diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11416 W: 2123 L: 2199 D: 7094
sprt @ 10+0.1 th 1 Plies To Promote idea...
09-02-16 My c=15 diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 12987 W: 2396 L: 2578 D: 8013
sprt @ 10+0.1 th 1 In light of tests http://tests.stockfishchess.org/tests/view/56b95fb10ebc590247cdfcdc http://tests.stockfishchess.org/tests/view/56b9dc790ebc590247cdfce8 Do we still have a regression on single thread? (Low priority *and throughput*)
09-02-16 Vo AP-LMR diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 68616 W: 12904 L: 12730 D: 42982
sprt @ 10+0.1 th 1 Decrease reduction if pawn advance to 7th rank.
09-02-16 jo lmrt1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 56740 W: 10697 L: 10655 D: 35388
sprt @ 10+0.1 th 1 Change reduction sequence in LMR. Test as parameter tweak.
09-02-16 SC qsearchVarianceRazor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5884 W: 1019 L: 1119 D: 3746
sprt @ 10+0.1 th 1 Last razoring attempt with factor = 2. Take 5 and then go to next search step.
09-02-16 Vo lmr-s diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 35670 W: 6595 L: 6816 D: 22259
sprt @ 10+0.1 th 1 LMR-Simplification
09-02-16 SC qsearchVarianceRazor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 27417 W: 5060 L: 5067 D: 17290
sprt @ 10+0.1 th 1 If several quiet moves are found, make possible to lower the value returned by qsearch by a something times the standard deviation of the qsearch values, in order to account for the uncertainity in the qsearch results. Take 3: use it in razoring with sigma/8.
09-02-16 My master diff
ELO: -0.50 +-2.8 (95%) LOS: 36.4%
Total: 20000 W: 3455 L: 3484 D: 13061
20000 @ 10+0.1 th 3 While empty I am curious to see what effect contempt now has with lazy SMP so quick measure (Low priority & fixed parsed variable)
09-02-16 sg capture_order diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16396 W: 3000 L: 3055 D: 10341
sprt @ 10+0.1 th 1 Try VoyagerOne's idea of sorting bad captures by SEE (see forum)
09-02-16 SC qsearchVarianceRazor diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 14377 W: 2633 L: 2697 D: 9047
sprt @ 10+0.1 th 1 Also use variance for fail highs and for storing to tt. Take 4: use it in razoring with sigma/16.
09-02-16 SC qsearchVarianceRazor diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 16285 W: 2966 L: 3022 D: 10297
sprt @ 10+0.1 th 1 If a several quiet moves are found, make possible to lower the value returned by qsearch by a something times the standard deviation of the qsearch values, in order to account for the uncertainity in the qsearch results. Take 2: use it in razoring with sigma.
09-02-16 pb variax diff
ELO: 3.75 +-5.5 (95%) LOS: 91.0%
Total: 5000 W: 841 L: 787 D: 3372
5000 @ 3+0.03 th 23 Smaller halfdensity-matrix with varied history bonus on second cycle. A local test with same conditions scored +6.39 ELO W: 924 - L: 832 - D: 3244 [0.509] 5000 but since I have 12 cores only I had to do it with Hyperthreading.
09-02-16 Ro Majority2 diff
LLR: -1.52 (-2.94,2.94) [0.00,5.00]
Total: 7331 W: 1326 L: 1359 D: 4646
sprt @ 10+0.1 th 1 Trying a much larger penalty (S(0, 40) instead of S(0, 8))
09-02-16 SC qsearchVarianceRazor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12924 W: 2299 L: 2369 D: 8256
sprt @ 10+0.1 th 1 If a several quiet moves are found, make possible to lower the value returned by qsearch by a something times the standard deviation of the qsearch values, in order to account for the uncertainity in the qsearch results. Take 1: use it in razoring with sigma/3.
09-02-16 My c=15 diff
ELO: 0.37 +-5.9 (95%) LOS: 54.9%
Total: 3724 W: 528 L: 524 D: 2672
20000 @ 30+0.3 th 3 Is there really no obvious ELO change from: http://tests.stockfishchess.org/tests/view/56b95fb10ebc590247cdfcdc Test how this scales at MTC, with hardcoded change to be certain. (Low priority)
09-02-16 Vo inCheck diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5388 W: 927 L: 1029 D: 3432
sprt @ 10+0.1 th 1 Don't update killers and cm if in check.
08-02-16 SC seeRefactoring diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 21959 W: 4089 L: 4119 D: 13751
sprt @ 10+0.1 th 1 Save some see calls, take 2.
08-02-16 Ro Majority2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17910 W: 3293 L: 3341 D: 11276
sprt @ 10+0.1 th 1 Fixed signature. Is there a problem with fishbench ?