Stockfish Testing Queue

Finished - 33316 tests

15-12-05 Roc BadPassedPawn diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 9864 W: 1948 L: 2031 D: 5885
sprt @ 10+0.1 th 1 When the opponent blocks the passed pawn and blocked sq not attacked by Us, reduce the passed pawn value as if one rank less.
15-12-05 Roc RookOnPawnThreat diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8478 W: 1587 L: 1676 D: 5215
sprt @ 10+0.1 th 1 Only consider threats by rook on queen, or on material not protected by pawns or minor.
15-12-03 Fis easyFix diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 40368 W: 5826 L: 5733 D: 28809
sprt @ 20+0.2 th 7 Bug fix: In case we make an easy move under SMP we can't select any other thread because ONLY the main thread has been verified to make it. Take 2 - no interaction with stillAtFirstMove logic. LTC
15-12-05 Voy rt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8832 W: 1656 L: 1744 D: 5432
sprt @ 10+0.1 th 1 Take 2
15-12-05 mbo tte2moves diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22050 W: 4216 L: 4245 D: 13589
sprt @ 10+0.1 th 1 Store two moves in TTE and use them for pruning and SE decisions. This also expands TTE key to 32-bits. (2nd try, updated with latest master)
15-12-05 Voy rt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10256 W: 1952 L: 2033 D: 6271
sprt @ 10+0.1 th 1 Refutation tweak...
15-12-05 Roc CowboysVsAliens diff
LLR: -2.08 (-2.94,2.94) [0.00,5.00]
Total: 3128 W: 583 L: 659 D: 1886
sprt @ 10+0.1 th 1 When minor(s) are attacking a piece defended only by one Major
15-12-05 Fis maxDepth diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 25267 W: 4762 L: 4966 D: 15539
sprt @ 10+0.1 th 1 Increase MAX_PLY from 128 to 250 regression test as requested.
15-12-05 jos no_space diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20490 W: 3940 L: 3975 D: 12575
sprt @ 10+0.1 th 1 Take 2.
15-12-05 mbo ponder_off_sm_92 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11368 W: 2104 L: 2181 D: 7083
sprt @ 10+0.1 th 1 Ponder=off, Slow Mover=92.
15-12-05 jos no_space^ diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12563 W: 2407 L: 2478 D: 7678
sprt @ 10+0.1 th 1 No space evaluation after move 29. Take 1.
15-12-04 mbo ponder_off diff
19877/20000 iterations
40000/40000 games played
40000 @ 10+0.1 th 1 Tune Slow Mover with ponder set to false. This is my first SPSA tuning session, please check that everything is correct before approving.
15-12-05 Voy rcm3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3340 W: 574 L: 686 D: 2080
sprt @ 10+0.1 th 1 One more shot at this...
15-12-04 pb0 distinct_iter_paths2 diff
ELO: -0.00 +-3.4 (95%) LOS: 50.0%
Total: 10000 W: 1277 L: 1277 D: 7446
10000 @ 5+0.1 th 7 Quick check if this simple lazy smp-formula is worth something. Now with 7 threads insted of 3, shorter TC and enhanced logic (provided with more skip-slots). To give it free as soon test with 3 threads finishes (distinct_iter_paths (easy approach)).
15-12-04 Voy rcm2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6777 W: 1280 L: 1377 D: 4120
sprt @ 10+0.1 th 1 Take 2.
15-12-04 Voy rcm diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25337 W: 4863 L: 4877 D: 15597
sprt @ 10+0.1 th 1 refutation criteria modifications.
15-12-04 n_p ThreatWeight diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 22718 W: 4499 L: 4259 D: 13960
sprt @ 10+0.1 th 1 Is pushing the MG value of threat weight higher, even better?
15-12-04 Roc B2B diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15376 W: 2975 L: 3033 D: 9368
sprt @ 10+0.1 th 1 Bishop on pawn experiment
15-12-04 Voy TH-LMR diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 4564 W: 813 L: 920 D: 2831
sprt @ 10+0.1 th 1 Take 2.
15-12-03 pb0 distinct_iter_paths diff
ELO: 1.04 +-3.5 (95%) LOS: 71.8%
Total: 10000 W: 1370 L: 1340 D: 7290
10000 @ 10+0.1 th 3 Quick check if this simple lazy smp-formula is worth something.
15-12-03 Voy TH-LMR diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 15655 W: 2987 L: 3044 D: 9624
sprt @ 10+0.1 th 1 Try using % of best H/CMH moves to assist with LMR.
15-12-03 Fis easyFix diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 27227 W: 4910 L: 4800 D: 17517
sprt @ 5+0.1 th 7 Bug fix: In case we make an easy move under SMP we can't select any other thread because ONLY the main thread has been verified to make it. Take 2 - no interaction with stillAtFirstMove logic.
15-12-03 sni outpost15 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 9314 W: 1786 L: 1871 D: 5657
sprt @ 10+0.1 th 1 Bonus for reachable outpost even for piece already in an outpost. Tested against master with green absimaldata's weights.
15-12-03 sni outpost15 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 5781 W: 1087 L: 1189 D: 3505
sprt @ 10+0.1 th 1 Double bonus for multiple reachable outposts. Tested against master with green absimaldata's weights.
15-12-03 mbo flat_lazy diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3697 W: 553 L: 661 D: 2483
sprt @ 10+0.1 th 3 Starts iterations with deepest completed depth plus one, and removes the score test from best thread selection.
15-12-03 jos big_tuning_depth1 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 3372 W: 589 L: 736 D: 2047
sprt @ 10+0.1 th 1 Tuning session 2. Test pawn values first.
15-12-02 mco TunedWeights diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 30157 W: 4540 L: 4303 D: 21314
sprt @ 60+0.4 th 1 LTC: Tuned evaluation weights.
15-12-02 IIv PV_vs_nonPV diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 34107 W: 6511 L: 6548 D: 21048
sprt @ 10+0.1 th 1 I want to see can I get a better approximation with a function different than log.
15-12-02 IIv PV_vs_nonPV diff
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 36331 W: 7117 L: 6829 D: 22385
sprt @ 10+0.1 th 1 Trying the average between master and its neutral counterpart. This is one of my last tries and I'll explain on the forum why I find this important.
15-12-02 Roc OutpostOnKing diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11968 W: 2298 L: 2372 D: 7298
sprt @ 10+0.1 th 1 Take 2: Smaller weights... and win ?!
15-12-01 pb0 slipstream_evase2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 18275 W: 3074 L: 3124 D: 12077
sprt @ 5+0.1 th 7 slipstream-evasion attempt nr. 2 with refined slipstream detection and evasion by ONE_PLY extension at depth 9
15-11-28 luc addDepthByBF diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 37803 W: 5389 L: 5376 D: 27038
sprt @ 20+0.2 th 7 As this looks highly TC-dependent and all the attempts went ‘yellow’ at STC, I'd like to test the simplest patch at longer TC. Low throughput (but stop if unappropriate).
15-12-02 jos big_tuning_depth1 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 1414 W: 256 L: 418 D: 740
sprt @ 10+0.1 th 1 I feel this will be a big disaster! This looks like the same strange values I got with Texel's tuning method. For some unknown reason this simply doesn't work for Stockfish.
15-12-01 adp TunedWeights diff
LLR: 1.32 (-2.94,2.94) [0.00,4.00]
Total: 190043 W: 37433 L: 36675 D: 115935
sprt @ 10+0.1 th 1 Tuned evaluation weights.
15-12-02 jos no_altering diff
ELO: -8.62 +-3.5 (95%) LOS: 0.0%
Total: 10000 W: 1194 L: 1442 D: 7364
10000 @ 10+0.1 th 3 Don't alter the search depth of the helpers inside the id_loop. This is most likely causing problems. A quick check to see how much we eventually lose. (The other option is to start and stop the threads with each iteration.)
15-11-30 Fis easyFix diff
LLR: 0.32 (-2.94,2.94) [-3.00,1.00]
Total: 18016 W: 2533 L: 2542 D: 12941
sprt @ 20+0.2 th 7 Bug fix: In case we make an easy move under SMP we can't select any other thread because ONLY the main thread has been verified to make it. LTC
15-12-02 sg pawns diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 3773 W: 728 L: 876 D: 2169
sprt @ 10+0.1 th 1 My try on fixed depth=1 SPSA tuning. Local tuned pawns.cpp (without king safety stuff) with 1M games.
15-12-01 mco fix_multipv diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 9759 W: 1734 L: 1595 D: 6430
sprt @ 5+0.1 th 7 Another attempt at fixing broken multi PV and skill levels.
15-12-02 Roc OutpostOnKing diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 16989 W: 3262 L: 3313 D: 10414
sprt @ 10+0.1 th 1 ...and win.
15-12-01 IIv PV_vs_nonPV diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 43379 W: 8288 L: 8292 D: 26799
sprt @ 10+0.1 th 1 Trying the opposite - decreasing the difference between PV and nonPV with depth.
15-12-01 Voy sa2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 11368 W: 2115 L: 2233 D: 7020
sprt @ 10+0.1 th 1 Take 2...
15-12-01 Voy sa diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 11706 W: 2182 L: 2299 D: 7225
sprt @ 10+0.1 th 1 Stats adjustment
15-12-01 lbr pawn diff
LLR: -2.94 (-2.94,2.94) [0.00,4.00]
Total: 8369 W: 1608 L: 1737 D: 5024
sprt @ 10+0.1 th 1 tuned at depth=1, 200k games
15-11-29 adp WeightTune diff
40122/40000 iterations
80000/80000 games played
80000 @ 20+0.2 th 1 Tuning of weights. I think there would be better weights due to recent changed in search. With added additional weight of Threats {350,256} as proposed by snicolet, as per request of a few people.
15-12-01 Roc WhenTheQueenIsGone diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 7354 W: 1351 L: 1445 D: 4558
sprt @ 10+0.1 th 1 ...do we still need the specialized rook eval terms ?
15-12-01 n_p Tuned_StormDanger diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 37861 W: 7323 L: 7346 D: 23192
sprt @ 10+0.1 th 1 Test of tuned StormDanger values obtained by SPSA.
15-12-01 cib avgscore diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20311 W: 3779 L: 3816 D: 12716
sprt @ 05+0.1 th 3 Take 2, using weighted average.
15-11-29 n_p Tune_StormDanger diff
49273/50000 iterations
100000/100000 games played
100000 @ 20+0.2 th 1 SPSA-session on StormDanger. Included tuned threat weights.
15-12-01 Voy ss diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8361 W: 1561 L: 1651 D: 5149
sprt @ 10+0.1 th 1 Don't update stats if reduced search is less than half of the depth of the current iteration.
15-11-29 Roc MobilityPawnTweak diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 25770 W: 5049 L: 5115 D: 15606
sprt @ 10+0.1 th 1 Amongst the rank2-3 pawns, only the center pawns are viewed as mobility blocker for our pieces. Run as "parameter tweak" test.