Stockfish Testing Queue

Finished - 30508 tests

16-07-02 SC branchingFactor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 49730 W: 9383 L: 9291 D: 31056
sprt @ 10+0.1 th 1 Retesting after implementing the actual idea proposed by DM.
16-07-03 IIv tmm_new diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 19435 W: 3502 L: 3693 D: 12240
sprt @ 10+0.1 th 1 I'm very close to the end of my work on time management. So, SPRT to see a progress or regression.
16-07-04 Voy shellSort diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9259 W: 1658 L: 1744 D: 5857
sprt @ 10+0.1 th 1 Shell sort seems to be > 4% faster than insertion. Bench is difference since shell is an unstable sort.
16-07-01 pb0 lazy_high_density diff
ELO: 2.43 +-3.6 (95%) LOS: 90.5%
Total: 12000 W: 2091 L: 2007 D: 7902
12000 @ 3+0.03 th 15 Retrying a high-density approach. (http://tests.stockfishchess.org/tests/view/567667fd0ebc592d552a42be might have been stopped to early)
16-07-01 pb0 lazy_high_density diff
ELO: 1.10 +-3.8 (95%) LOS: 71.7%
Total: 12000 W: 2214 L: 2176 D: 7610
12000 @ 3+0.03 th 7 Retrying a high-density approach. Take on 7 threads
16-07-02 IIv tmm_new diff
ELO: 9.21 +-4.3 (95%) LOS: 100.0%
Total: 10000 W: 2175 L: 1910 D: 5915
10000 @ 15+0 th 1 Measure sudden death Elo performance of my newest version of time management.
16-07-02 sni komodo3' diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17645 W: 3265 L: 3314 D: 11066
sprt @ 10+0.1 th 1 Try tuned values after 51.000 games
16-07-02 cib stats_reorg diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 17683 W: 3325 L: 3374 D: 10984
sprt @ 10+0.1 th 1 Take 2.
16-07-02 sni hanging diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13148 W: 2403 L: 2472 D: 8273
sprt @ 10+0.1 th 1 Double the bonus when the side to move can capture a hanging enemy.
16-07-02 SC LMRstats diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6270 W: 1117 L: 1216 D: 3937
sprt @ 10+0.1 th 1 Tuned values.
16-06-24 sg quience_search diff
LLR: -1.93 (-2.94,2.94) [-3.00,1.00]
Total: 120395 W: 22623 L: 22939 D: 74833
sprt @ 10+0.1 th 1 remove recapture phase from quience search move generation
16-06-29 Roc MoreCheck diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 54550 W: 7671 L: 7365 D: 39514
sprt @ 60+0.6 th 1 LTC for this king safety patch which was yellow at STC
16-06-26 SC LMRstatsTuning diff
46761/50000 iterations
94320/100000 games played
100000 @ 10+0.1 th 1 I dont have the slightest idea of whether the parameters I chose are ok, but it was not that bad. So try to tune it. Restart with higher c. Will stop if it converges early.
16-07-01 sg king_safety diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 56342 W: 10947 L: 10822 D: 34573
sprt @ 10+0.1 th 1 Recognize pieces which defend attacked squares in the king ring. Take 2
16-06-27 SC futilityPawns diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 143212 W: 19630 L: 19676 D: 103906
sprt @ 60+0.6 th 1 Prune futile nodes also if we have only pawns. Was +2 ELO after 1000 local games, so for SPRT. LTC. Fixed hash.
16-07-02 sni komodo3' diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12022 W: 2184 L: 2258 D: 7580
sprt @ 10+0.1 th 1 Try half the values of tuning after 37k games
16-07-01 Voy trueCM2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20159 W: 3730 L: 3768 D: 12661
sprt @ 10+0.1 th 1 Take 2 (fix).
16-07-02 cib stats_reorg diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12184 W: 2231 L: 2304 D: 7649
sprt @ 10+0.1 th 1 Rearranging the update stats code. Making behaviour consistent between TT hit, and non TT hit updates.
16-07-01 Roc TacticalKingRing diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 28722 W: 5411 L: 5411 D: 17900
sprt @ 10+0.1 th 1 Always scored the tactical threats involving king,
16-07-01 Voy trueCM2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6655 W: 1186 L: 1283 D: 4186
sprt @ 10+0.1 th 1 Take 2. Using 2 trueCMs.
16-07-01 sni tropism2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17957 W: 3423 L: 3470 D: 11064
sprt @ 10+0.1 th 1 King tropism in endgame too
16-07-01 SC initiativeEqual diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 4698 W: 894 L: 1001 D: 2803
sprt @ 10+0.1 th 1 Maybe more interesting than contempt: initiative in equal positions.
16-06-30 Fis splitMobilityTune diff
13458/25000 iterations
27186/50000 games played
50000 @ 20+0.2 th 1 Split the mobility board into two regions roughly similar to what PSQT considers good or bad. Tune.
16-07-01 sni komodo3 diff
ELO: -3.31 +-3.4 (95%) LOS: 2.6%
Total: 15725 W: 2938 L: 3088 D: 9699
20000 @ 10+0.1 th 1 Different strategies depending if SF is winning or losing. Take 1: only change the strategy when SF is losing.
16-06-24 pb0 halfdensity_8block diff
ELO: -6.11 +-3.5 (95%) LOS: 0.0%
Total: 12000 W: 1831 L: 2042 D: 8127
20000 @ 3+0.03 th 21 Agreed with Peter Zsifkovits (CoffeeOne) to do some tests on large number of threads. First one is to verify if the last block in the halfdensitymap (skipsize=4) brings any benefit.
16-07-01 pb0 lazy_mixed_map diff
ELO: -5.48 +-5.5 (95%) LOS: 2.6%
Total: 5259 W: 873 L: 956 D: 3430
20000 @ 3+0.03 th 7 As proposed by Leonid Pechenik trying intermittent 1-2 skips before skip size 2 patterns
16-06-30 jos tm-tweak diff
ELO: 1.13 +-4.1 (95%) LOS: 70.3%
Total: 8000 W: 1198 L: 1172 D: 5630
8000 @ 40+0.4 th 1 Just a quick check if there is any gain at longer tc.
16-06-30 Mys lp diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 38592 W: 7273 L: 7230 D: 24089
sprt @ 10+0.1 th 1 Try a small change to loose enemies condition
16-06-29 Voy trueCM diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 47548 W: 8947 L: 8865 D: 29736
sprt @ 10+0.1 th 1 Define piece in cm.
16-06-28 jos tuneTMM diff
13327/20000 iterations
26968/40000 games played
40000 @ 40+0.4 th 1 Retune time management at 4 x longer tc. SF sometimes moves too fast, imho. Do we get different values? (Half throughput.)
16-06-29 Elb end_draw diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 39415 W: 5870 L: 5609 D: 27936
sprt @ 10+0.1 th 1 My take on drawish endgames with equal material and compact pawn chains.
16-06-28 pb0 circular_clear_cm diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 12736 W: 2053 L: 2126 D: 8557
sprt @ 5+0.05 th 7 STC: Clear countermoves of helper-threads with a round robin
16-06-29 sg king_safety diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 28491 W: 5561 L: 5560 D: 17370
sprt @ 10+0.1 th 1 Recognize pieces which defend attacked squares in the king ring
16-06-29 SC branchingFactor diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15564 W: 2866 L: 2924 D: 9774
sprt @ 10+0.1 th 1 Implementation of idea by Dragon Mist in the forum https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/WOXnuS8DuMc
16-06-29 Fis splitMobilityTune diff
12956/20000 iterations
26686/40000 games played
40000 @ 20+0.2 th 1 Split mobility into lower and upper ranks. Tune.
16-06-29 jos tm-tweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 11456 W: 2130 L: 2248 D: 7078
sprt @ 10+0.1 th 1 Test the tuned values. (Not quite as expected, but MIN is slightly increased, MAX stayed the same, easy move time is significantly reduced, and the time plus for fail-lows increased.)
16-06-29 sni tune_komodo diff
14837/15000 iterations
30000/30000 games played
30000 @ 10+0.1 th 1 Tune with fixed optimism_mobility=10
16-06-29 sni komodo diff
ELO: -1.84 +-3.1 (95%) LOS: 12.1%
Total: 20000 W: 4065 L: 4171 D: 11764
20000 @ 10+0.1 th 1 Take 5bis: half values of take 5
16-06-29 SC ZeroDepthPruning diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 7095 W: 1270 L: 1441 D: 4384
sprt @ 10+0.1 th 1 A variation of the previous ParentNodePruning idea. Will see whether this or the previous one is better, tune it and then have a last go.
16-06-28 sni komodo diff
ELO: -2.55 +-3.2 (95%) LOS: 5.9%
Total: 20000 W: 4334 L: 4481 D: 11185
20000 @ 10+0.1 th 1 Take 5: try the resulting values of tuning #3 (with fixed OPTIMISM_PAWNS=10 for us)
16-06-29 cib prior_killer diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17378 W: 3225 L: 3275 D: 10878
sprt @ 10+0.1 th 1 Updating killer moves after fail low.
16-06-28 sni tropism2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19752 W: 3736 L: 3775 D: 12241
sprt @ 10+0.1 th 1 King tropism using double attacks, take 2.
16-06-28 Voy clearRCM diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17334 W: 3248 L: 3298 D: 10788
sprt @ 10+0.1 th 1 Clear refuted cm.
16-06-28 SC ParentNodePruning diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 15553 W: 2834 L: 3019 D: 9700
sprt @ 10+0.1 th 1 Is it possible to unify razoring and futility pruning? An exotic, locally tuned attempt.
16-06-28 sni komodo diff
ELO: 2.26 +-2.9 (95%) LOS: 93.4%
Total: 20000 W: 3801 L: 3671 D: 12528
20000 @ 10+0.1 th 1 Take 4bis: half values of take 4
16-06-28 sni komodo diff
ELO: -0.47 +-2.9 (95%) LOS: 37.6%
Total: 20000 W: 3641 L: 3668 D: 12691
20000 @ 10+0.1 th 1 Take 4: try the resulting value of tuning #2 (with fixed OPTIMISM_PIECES=10 for us)
16-06-27 sni tune_komodo diff
14772/15000 iterations
30000/30000 games played
30000 @ 10+0.1 th 1 Tune with fixed optimism_pawns=10
16-06-28 pb0 circular_clear_cm diff
ELO: 4.04 +-4.6 (95%) LOS: 95.9%
Total: 8000 W: 1485 L: 1392 D: 5123
8000 @ 3+0.03 th 7 Clear countermoves of helper-threads with a round robin
16-06-28 Mys ms diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8893 W: 1622 L: 1709 D: 5562
sprt @ 10+0.1 th 1 cap opp bishops material
16-06-25 pb0 lazy_big_map diff
ELO: 0.59 +-3.7 (95%) LOS: 62.2%
Total: 10000 W: 1504 L: 1487 D: 7009
10000 @ 3+0.03 th 44 Can we further extend the halfdensity map? Quick check.