Stockfish Testing Queue

Pending - 0 tests 0.0 hrs

None

Active - 0 tests

Finished - 339 tests

24-03-17 pb checkExt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 36845 W: 6553 L: 6523 D: 23769
sprt @ 10+0.1 th 1 Retire check extension on late evasion move when checking move was late too (bugfix). Test on top of passed extT.
24-03-17 pb checkExt diff
LLR: -0.19 (-2.94,2.94) [0.00,5.00]
Total: 513 W: 97 L: 103 D: 313
sprt @ 10+0.1 th 1 Retire check extension on late evasion move when checking move was late too. Test on top of passed extT
23-03-17 pb singular_checking diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 44595 W: 7999 L: 7936 D: 28660
sprt @ 10+0.1 th 1 Singular Extension search: search one ply more when checking. Test on top of passed extT
22-03-17 pb killerCapture diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 37244 W: 6635 L: 6603 D: 24006
sprt @ 10+0.1 th 1 Sophisticated capturekiller logic showing good local results.
20-03-17 pb advanced_pawn_push diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 27201 W: 4832 L: 4842 D: 17527
sprt @ 10+0.1 th 1 Refine definition of advanced pawn push
18-03-17 pb verification_search8 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18657 W: 1946 L: 2010 D: 14701
sprt @ 30+0.3 th 1 Does the zugzang-awareness allow us to null-move-prune also in pv-nodes? STC with 3 x times larger TC (1/3 throughput) to hit modified verification search often enough.
16-03-17 pb verification_search8 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 33809 W: 3574 L: 3597 D: 26638
sprt @ 30+0.3 th 1 Trying yet another variant: STC with 3 times larger TC (1/3 throughput) to trigger modified verification search enough often to measure any difference.
15-03-17 pb verification_search4'' diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 45367 W: 4088 L: 4094 D: 37185
sprt @ 60+0.6 th 1 LTC: My try on make SF more aware of zugzwang, take 2: now verification triggered again not before depth 12. (8 moves book)
15-03-17 pb verification_search4'' diff
ELO: 2.35 +-2.1 (95%) LOS: 98.8%
Total: 19660 W: 1824 L: 1691 D: 16145
25000 @ 60+0.6 th 1 My try on make SF more aware of zugzwang, take 2: now with triggered again not before depth 12. (lower throughput and 8 moves book)
14-03-17 pb verification_search4' diff
ELO: 1.51 +-2.1 (95%) LOS: 92.3%
Total: 19561 W: 1825 L: 1740 D: 15996
20000 @ 60+0.6 th 1 My try on make SF more aware of zugzwang based on verification search. On pos FEN: 1k3b1q/pP2p1p1/P1K1P1Pp/7P/2B5/8/8/8 w - - 0 1 it finds Bb5 in 11 secs. (lower throughput.)
14-03-17 pb strikeBack diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 12691 W: 2219 L: 2291 D: 8181
sprt @ 10+0.1 th 1 Take 3
14-03-17 pb strikeBack diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 28139 W: 4977 L: 4984 D: 18178
sprt @ 10+0.1 th 1 Take 2: more accurate & double hits by capture escape detection
13-03-17 pb strikeBack diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 29151 W: 5168 L: 5170 D: 18813
sprt @ 10+0.1 th 1 Another exotic experiment on the idea of using an alternative strikeBack square for see.
13-03-17 pb capture_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12540 W: 2185 L: 2257 D: 8098
sprt @ 10+0.1 th 1 One more shot at this idea...
13-03-17 pb verification_search4' diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 11107 W: 1683 L: 1763 D: 7661
sprt @ 20+0.2 th 1 Respin this zugzwang-aware 100k yellow http://tests.stockfishchess.org/tests/view/584578800ebc5903140c5704 (rebased) at 20+0.2 (Throughout x1/2) since verification search might be tc sensitive.
09-03-17 pb skipQuiets2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12445 W: 2146 L: 2219 D: 8080
sprt @ 10+0.1 th 1 Combine skipQuiets with futility pruning. Test on top of passed skipQuiets' patch.
11-03-17 pb anticipate_mcpruning diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 14558 W: 2581 L: 2644 D: 9333
sprt @ 10+0.1 th 1 Anticipate MoveCountPuning when conditions for parent node futility pruning apply (based off and against skipQuiets')
11-03-17 pb skipQuiets2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6419 W: 1133 L: 1231 D: 4055
sprt @ 10+0.1 th 1 Take 7: exclude direct checks and adv. pawn pushes from being skipped. (based off and against skipQuiets')
10-03-17 pb skipQuiets2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 71174 W: 12888 L: 12711 D: 45575
sprt @ 10+0.1 th 1 Take 5: exclude checking moves from being skipped (fixed bench)
10-03-17 pb skipQuiets3 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 10704 W: 1901 L: 1981 D: 6822
sprt @ 10+0.1 th 1 Take 6: Exclude checks as well as advanced pawn pushes from being skipped. This should allow us to trigger skipQuiets also on futility pruning parent node (test still against passed skipQuiets' patch)
09-03-17 pb skipQuiets2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 11247 W: 1956 L: 2034 D: 7257
sprt @ 10+0.1 th 1 Take 4: (Test on top of passed skipQuiets')
09-03-17 pb skipQuiets2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 1619 W: 229 L: 347 D: 1043
sprt @ 10+0.1 th 1 Take 3: Suggestion by VoyagerOne. (Test on top of passed skipQuiets')
09-03-17 pb skipQuiets2 diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 25320 W: 4565 L: 4582 D: 16173
sprt @ 10+0.1 th 1 Take 2: A more sensible approach. Test on top of passed skipQuiets' (fixed base bench)
08-03-17 pb one_bit_in_tt diff
ELO: -3.01 +-2.9 (95%) LOS: 1.9%
Total: 20000 W: 3432 L: 3605 D: 12963
20000 @ 10+0.1 th 1 Take 2: how many ELO we would loose by detecting positions with hanging pieces and merely store it as bool info in TT (without exploiting it)
07-03-17 pb low_depth_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 24771 W: 4355 L: 4428 D: 15988
sprt @ 10+0.1 th 1 Take 4
07-03-17 pb low_depth_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 38393 W: 6731 L: 6759 D: 24903
sprt @ 10+0.1 th 1 Take 3: since first 2 takes were quick-fails, try to increase this factor which I think wasn't yet re-tuned after #788
07-03-17 pb low_depth_pruning diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10106 W: 1757 L: 1839 D: 6510
sprt @ 10+0.1 th 1 Take 2
06-03-17 pb one_bit_in_tt diff
ELO: -0.86 +-2.6 (95%) LOS: 25.7%
Total: 25000 W: 4475 L: 4537 D: 15988
25000 @ 10+0.1 th 1 Would like to measure how many ELO we would loose by sacrificing one bit in TT-generation for other general purpose (e.g. marking positions with hanging pieces)
05-03-17 pb low_depth_pruning diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8790 W: 1533 L: 1621 D: 5636
sprt @ 10+0.1 th 1 Combine futility pruning parent node & see-based pruning
04-03-17 pb see_advanced_pawn diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14733 W: 2603 L: 2666 D: 9464
sprt @ 10+0.1 th 1 More value for advanced pawn in SEE. I believe this was'nt tried since we see_prune at shallow depth
03-03-17 pb bad_good_2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 41992 W: 7554 L: 7501 D: 26937
sprt @ 10+0.1 th 1 Take 4: When a bad-capture becomes bestMove, use it as good-capture in sibling nodes (only real siblings nodes this time)
01-03-17 pb xray_on_pinned diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18340 W: 3242 L: 3290 D: 11808
sprt @ 10+0.1 th 1 Like to see it this idea has any value...
01-03-17 pb bad_good_2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 11082 W: 1897 L: 1975 D: 7210
sprt @ 10+0.1 th 1 Take 3: Very similar to take 1 but include also early TT cutoffs
01-03-17 pb bad_good_2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 14436 W: 2587 L: 2651 D: 9198
sprt @ 10+0.1 th 1 Capture killer, include tt-hits, distinct bad/good based on see-value (bugfix, previous patch movepicker returned moves twice)
01-03-17 pb bad_good_2 diff
LLR: -1.72 (-2.94,2.94) [0.00,5.00]
Total: 4162 W: 713 L: 768 D: 2681
sprt @ 10+0.1 th 1 Capture killer before good-captures, distinct bad/good based on see-value, set killer also on early TT cutoffs.
28-02-17 pb bad_good_2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 36629 W: 6419 L: 6391 D: 23819
sprt @ 10+0.1 th 1 When a bad-capture becomes bestMove, use it as good-capture in sibling nodes.
27-02-17 pb capture_history diff
LLR: -5.01 (-2.94,2.94) [0.00,5.00]
Total: 32779 W: 5759 L: 5833 D: 21187
sprt @ 10+0.1 th 1 Take 6
27-02-17 pb capture_history diff
LLR: -2.92 (-2.94,2.94) [0.00,5.00]
Total: 4748 W: 832 L: 936 D: 2980
sprt @ 10+0.1 th 1 Take 5
27-02-17 pb stat_update diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 739 W: 84 L: 206 D: 449
sprt @ 10+0.1 th 1 Update statistics for every quiet move that raises alpha, so that quietsSearched is'nt needed anymore.
27-02-17 pb capture_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13464 W: 2343 L: 2411 D: 8710
sprt @ 10+0.1 th 1 Take 4: Avoid with ca. 78% accuracy see-call for captures with non-negative see-value
24-02-17 pb capture_history diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 22858 W: 4058 L: 4087 D: 14713
sprt @ 10+0.1 th 1 Take 3: don't call SEE on captures with positive history value
24-02-17 pb capture_history2 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 997 W: 123 L: 244 D: 630
sprt @ 10+0.1 th 1 Take 4: Decide bad/good capture based on history value
23-02-17 pb capture_history diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 40522 W: 7228 L: 7182 D: 26112
sprt @ 10+0.1 th 1 Take 2 (fixed bench)
22-02-17 pb castling_in_see2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 48310 W: 8692 L: 8684 D: 30934
sprt @ 10+0.1 th 1 Take 3 (last try): Moving castling condition from see_ge to qsearch futility pruning in way that castling-moves never get pruned.
21-02-17 pb castling_in_see2 diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 25956 W: 4504 L: 4573 D: 16879
sprt @ 10+0.1 th 1 Obtaining a non-functional-patch by moving castling condition from see_ge to qsearch futility pruning where it get invoked less often. Benchmark shows no significant speed diff. but we might have less castling moves in a real game in average than in a bench run.
20-02-17 pb capture_history diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 21184 W: 3769 L: 3805 D: 13610
sprt @ 10+0.1 th 1 Another wild experiment for a capture heuristic
20-02-17 pb castling diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 60833 W: 10932 L: 10882 D: 39019
sprt @ 10+0.1 th 1 Testing as improvement ([0,4] bounds) a variant of Ronald de Mans alternative encoding of castling. For more info, see comments on https://github.com/pb00068/Stockfish/commit/b10b0f6b37f14c69d2d0007210049f4b1e8be4d7
20-02-17 pb capture_escape5 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 15150 W: 2698 L: 2759 D: 9693
sprt @ 10+0.1 th 1 Want to measure the isolated effect of applying escape-capture-logic on capture-moves too.
20-02-17 pb recaptures diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 9092 W: 1635 L: 1722 D: 5735
sprt @ 10+0.1 th 1 Another (probably last) shot at this idea.
17-02-17 pb recaptures diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 21323 W: 3811 L: 3846 D: 13666
sprt @ 10+0.1 th 1 Bonus instead of Malus. Running test for snicolet.