Stockfish Testing Queue

Finished - 40849 tests

17-02-22 Gua QDTunes diff
13216/15000 iterations
29834/30000 games played
30000 @ 5+0.05 th 1 will it progress (fast tunes)
17-02-22 sni definition_of_weak3 diff
LLR: 2.97 (-2.94,2.94) [0.00,5.00]
Total: 17050 W: 3128 L: 2931 D: 10991
sprt @ 10+0.1 th 1 Change definition 'weak' in threats calculation
17-02-22 sni pawn_chains8 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 20790 W: 3750 L: 3787 D: 13253
sprt @ 10+0.1 th 1 Take 2
17-02-18 vdv skipsf diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 84255 W: 12457 L: 12731 D: 59067
sprt @ 5+0.05 th 11 test simplification of skip_ply, which is different on 11 threads (but not e.g. 7).
17-02-22 Gua KDTunes diff
498/25000 iterations
5018/50000 games played
50000 @ 1+0.01 th 1 KingDistanceTunes (try very fast)
17-02-22 sg tune_four_phases_psqt_p diff
2866/50000 iterations
6004/100000 games played
100000 @ 20+0.2 th 1 The tuned values are not good. Repeat the tuning now with lower ck=5 instead of ck=10 to.
17-02-22 Elb nmpt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 7358 W: 1268 L: 1362 D: 4728
sprt @ 10+0.1 th 1 NMP and futility pruning
17-02-22 sg four_phases_tuned_psqt_ diff
ELO: -3.09 +-4.5 (95%) LOS: 9.1%
Total: 10328 W: 2339 L: 2431 D: 5558
20000 @ 10+0.1 th 1 Quick test of the first four phases tuning. Some parameters seems strange but many have clear trends. perhaps ck=10 is to high and a further tuning is necessary. But lets look at the result.
17-02-22 Gua QDistance-1 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 5139 W: 861 L: 964 D: 3314
sprt @ 10+0.1 th 1 Bonuses, depending on queen distance
17-02-22 sni pawn_chains8 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20875 W: 3693 L: 3730 D: 13452
sprt @ 10+0.1 th 1 Central pawn chains
17-02-21 sg tune_four_phases_psqt_p diff
42857/50000 iterations
93472/100000 games played
100000 @ 20+0.2 th 1 I extended the evaluation from two to four phases: opening, middlegame, endgame, late endgame. First try tuning the pawn piece square table (96 parameters). Retry the tuning with fixed version (Thanks to Ivan Ivec). Added last parameter (which should be always zero) to trigger UPDATE_LAST_ON:
17-02-22 pb0 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.
17-02-22 Elb nmpt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 6932 W: 1191 L: 1287 D: 4454
sprt @ 10+0.1 th 1 Don't return nullValue too soon
17-02-22 sni loose_enemies diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12843 W: 1680 L: 1756 D: 9407
sprt @ 60+0.6 th 1 LTC: Endgame bonus if the opponent has loose pieces
17-02-21 SC evalDoubleChecks diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 67347 W: 12188 L: 12027 D: 43132
sprt @ 10+0.1 th 1 Always extend double checks, inspired by sg.
17-02-22 sni loose_enemies diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 67534 W: 12292 L: 11881 D: 43361
sprt @ 10+0.1 th 1 Endgame bonus if the opponent has loose pieces
17-02-22 Voy stackHistory diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25765 W: 4588 L: 4604 D: 16573
sprt @ 10+0.1 th 1 reset history after singular
17-02-21 Voy stackBugFix' diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 10568 W: 1813 L: 1987 D: 6768
sprt @ 10+0.1 th 1 Based on conversation with loco, stack information will be wrong when searching at the same stack level (i.e. singular extension) so we should reinitialize some stack fields.
17-02-21 SC IID diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 30602 W: 5510 L: 5562 D: 19530
sprt @ 10+0.1 th 1 A different formula for IID
17-02-21 Roc XRayMob diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12242 W: 2186 L: 2259 D: 7797
sprt @ 10+0.1 th 1 And what if we simply exclude only the rook home rank xrays through queen ?
17-02-21 Anb movepick diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 4814 W: 794 L: 959 D: 3061
sprt @ 10+0.1 th 1 STC: Always partition quiet moves before sorting
17-02-21 sg tune_four_phases_psqt_p diff
422/50000 iterations
883/100000 games played
100000 @ 20+0.2 th 1 I extended the evaluation from two to four phases: opening, middlegame, endgame, late endgame. First try tuning the pawn piece square table (96 parameters). Retry the tuning with fixed version (Thanks to Ivan Ivec).
17-02-21 sg double_check diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6872 W: 1214 L: 1310 D: 4348
sprt @ 10+0.1 th 1 Extends this idea to discovered check
17-02-21 SC SES diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 5026 W: 840 L: 979 D: 3207
sprt @ 10+0.1 th 1 Another depth in SES.
17-02-21 SC simpleLMRcapture diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 1794 W: 233 L: 390 D: 1171
sprt @ 10+0.1 th 1 Handle captures exactly as all other moves, but reduce one additional ply.
17-02-21 pb0 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.
17-02-21 sg double_check2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8104 W: 1413 L: 1504 D: 5187
sprt @ 10+0.1 th 1 Less LMR if in double check.
17-02-21 sg double_check diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 44132 W: 7896 L: 7835 D: 28401
sprt @ 10+0.1 th 1 No pruning if in double check
17-02-21 sg double_check diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 21804 W: 3891 L: 3924 D: 13989
sprt @ 10+0.1 th 1 No LMR if in double check
17-02-21 Voy statTest diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 8889 W: 1570 L: 1657 D: 5662
sprt @ 10+0.1 th 1 Not sure why the bench will change...so I want to test this out.
17-02-21 Anb move-pick-tests diff
LLR: -1.00 (-2.94,2.94) [-3.00,1.00]
Total: 987 W: 146 L: 200 D: 641
sprt @ 10+0.1 th 1 STC: Possible to retire insertion_sort, use pick_best? Take1
17-02-20 fau knightdecmg diff
LLR: -2.87 (-2.94,2.94) [0.00,4.00]
Total: 67158 W: 12024 L: 11949 D: 43185
sprt @ 10+0.1 th 1 Knight psqt decrease -10% in mg, and 50% closer to 0 average.
17-02-21 Elb nmpt diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 13030 W: 2263 L: 2442 D: 8325
sprt @ 10+0.1 th 1 Use simpler null move dynamic reduction formula
17-02-20 Gua KingDistance diff
LLR: 0.11 (-2.94,2.94) [0.00,5.00]
Total: 9316 W: 1250 L: 1215 D: 6851
sprt @ 60+0.6 th 1 v4 was close, how it will on LTC
17-02-21 nhu initiative diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 32565 W: 5820 L: 5866 D: 20879
sprt @ 10+0.1 th 1
17-02-20 Voy singleKiller diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 64818 W: 11646 L: 11914 D: 41258
sprt @ 10+0.1 th 1 Test to see if we just need one killer move...
17-02-21 sg killer diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 6095 W: 1067 L: 1166 D: 3862
sprt @ 10+0.1 th 1 Use now only the first killer, which is way more often good (value > alpha) than the second killer.
17-02-21 Gua KDProtection diff
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 26177 W: 3642 L: 3435 D: 19100
sprt @ 60+0.6 th 1 (Protective bonuses, depending on the distance from the own King) LTC
17-02-20 fau knightdeceg diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 52731 W: 9488 L: 9465 D: 33778
sprt @ 10+0.1 th 1 Knight psqt decrease -10% in eg, and 50% closer to 0 average.
17-02-20 Roc XRayMob diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 59840 W: 10780 L: 10652 D: 38408
sprt @ 10+0.1 th 1 Take 4, suggested by snicolet, applied in the bishop case only.
17-02-21 sg lazy_eval diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 24655 W: 4371 L: 4392 D: 15892
sprt @ 10+0.1 th 1 Use only the scaling factor from material probe for lazy eval
17-02-21 sni entry_points4 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16996 W: 2992 L: 3045 D: 10959
sprt @ 10+0.1 th 1 Take 3
17-02-21 Fis flanks diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 37562 W: 6697 L: 6726 D: 24139
sprt @ 10+0.1 th 1 Don't count middle files in flanks.
17-02-20 sg tune_four_phases_psqt_p diff
19434/50000 iterations
39869/100000 games played
100000 @ 20+0.2 th 1 I extended the evaluation from two to four phases: opening, middlegame, endgame, late endgame. First try tuning the pawn piece square table (96 parameters). Later i post on the forum more informations.
17-02-20 Anb king-knight diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 43228 W: 7791 L: 7733 D: 27704
sprt @ 10+0.1 th 1 STC: use tuned mg values for closest knight to king
17-02-20 Gua KingDistance5 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 42001 W: 7752 L: 7696 D: 26553
sprt @ 10+0.1 th 1 v5
17-02-20 Voy mcpCapture diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 8068 W: 1405 L: 1496 D: 5167
sprt @ 10+0.1 th 1 stc
17-02-20 Elb nmpt diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17120 W: 3037 L: 3090 D: 10993
sprt @ 10+0.1 th 1 NMP tweak
17-02-20 Gua KDProtection diff
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 21192 W: 3919 L: 3704 D: 13569
sprt @ 10+0.1 th 1 Attack bonuses correlate with protective bonuses, but maybe it will better (try only protection bonuses v4)
17-02-20 sg lazy_eval diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19330 W: 3436 L: 3479 D: 12415
sprt @ 10+0.1 th 1 Try first full scaling factor for lazy eval.