Stockfish Testing Queue

Finished - 37472 tests

15-02-03 Roc LatentQAttack diff 
12179/12500 iterations

25000/25000 games played
25000 @ 15+0.05 th 1 SPSA run 1, starting at S(20,20) only 25M to find the trend if any
15-02-03 Roc LatentQAttack diff 
12264/12500 iterations

25001/25000 games played
25000 @ 15+0.05 th 1 SPSA run 2, starting at S(30,30), only 25M to find the trend if any.
15-02-03 sni pawn_support diff 
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]

Total: 16024 W: 3233 L: 3296 D: 9495
sprt @ 15+0.05 th 1 Bonus for safe pawn pushes getting connected pawns
15-02-03 gli movecount_qsearch diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 3971 W: 744 L: 840 D: 2387
sprt @ 15+0.05 th 1 One more try on making movecount pruning more accurate (hopefully cheaply enough this time)
15-02-03 sni piece_support diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 8297 W: 1559 L: 1643 D: 5095
sprt @ 15+0.05 th 1 Bonus for safe pawn pushes supporting one of our (centralized) pieces.
15-02-04 sg pawn_attack_threat3 diff 
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]

Total: 14594 W: 2876 L: 2943 D: 8775
sprt @ 15+0.05 th 1 Recognize only attacks on minor pieces (That was the original idea of Ludmil, i extended that in my succesful patch to all pieces).
15-02-04 Roc LatentQAttack diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 30549 W: 6060 L: 6083 D: 18406
sprt @ 15+0.05 th 1 See if we can trust the first SPSA result which was started with S(20,20) and roughly gave S(21,16) in all cases.
15-02-04 sni piece_support diff 
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]

Total: 4558 W: 903 L: 998 D: 2657
sprt @ 15+0.05 th 1 Take 2
15-02-04 sni pawn_support diff 
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]

Total: 13462 W: 2759 L: 2615 D: 8088
sprt @ 15+0.05 th 1 Take 2
15-02-04 sni pawn_attack_threat5 diff 
LLR: 4.25 (-2.94,2.94) [-1.50,4.50]

Total: 28734 W: 5846 L: 5613 D: 17275
sprt @ 15+0.05 th 1 Treat pawn pushes supported by pieces as safe pushes
15-02-04 jos matimb diff 
ELO: 1.84 +-2.2 (95%) LOS: 95.3%

Total: 40000 W: 8131 L: 7919 D: 23950
40000 @ 15+0.05 th 1 Check new values after 2nd SPSA session.
15-02-04 n_p KingSafety diff 
ELO: -1.08 +-2.2 (95%) LOS: 16.3%

Total: 40000 W: 7936 L: 8060 D: 24004
40000 @ 15+0.05 th 1 Checking the values from the SPSA-session on king safety.
15-02-04 SC tuned_tempo diff 
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]

Total: 9188 W: 1771 L: 1853 D: 5564
sprt @ 15+0.05 th 1 Increase tempo value in middlegame and decrease in endgame.
15-02-04 SC tuned_tempo diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 10206 W: 1974 L: 2053 D: 6179
sprt @ 15+0.05 th 1 Decrease tempo value for middlegame and increase for endgame.
15-02-04 sg pawn_attack_threat3 diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 15280 W: 3015 L: 3080 D: 9185
sprt @ 15+0.05 th 1 exclude queens as target
15-02-04 Roc LatentQAttack diff 
12145/12500 iterations

25000/25000 games played
25000 @ 15+0.05 th 1 SPSA run 1 looked ok, the test at SPRT 15 +0.05 was roughly 50/50 with quite a low draw rate. Another 25M games for more tuning.
15-02-04 lbr kmpkm diff 
LLR: 2.96 (-2.94,2.94) [-3.50,0.50]

Total: 21597 W: 3470 L: 3367 D: 14760
sprt @ 15+0.05 th 1 are KmPKm also useless?
15-02-04 sni pawn_support diff 
LLR: -2.95 (-2.94,2.94) [0.00,6.00]

Total: 29087 W: 4864 L: 4827 D: 19396
sprt @ 60+0.05 th 1 LTC: take 2
15-02-04 sni pawn_attack_threat5 diff 
LLR: -3.00 (-2.94,2.94) [0.00,4.00]

Total: 72097 W: 12026 L: 11952 D: 48119
sprt @ 60+0.05 th 1 LTC : treat pawn pushes supported by pieces as safe pushes. Tested with SPRT[0,4] as this is a tuning patch.
15-02-05 lbr kqkr diff 
LLR: 3.82 (-2.94,2.94) [-3.50,0.50]

Total: 72660 W: 11287 L: 11255 D: 50118
sprt @ 15+0.05 th 1 is KQKR useless ?
15-02-05 jos spsa_queen_imbal diff 
55624/50000 iterations

107000/100000 games played
100000 @ 15+0.05 th 1 Try a 3rd tuning session with the values from the 2nd one as starting point. Now with smaller ck.
15-02-05 n_p KingSafety diff 
LLR: -2.96 (-2.94,2.94) [0.00,4.00]

Total: 16384 W: 2594 L: 2696 D: 11094
sprt @ 60+0.05 th 1 There does not seem to be a significant gain in STC but because king safety is TC dependent and tuning done in LTC, chech if improvement in LTC.
15-02-05 sni connected_passed diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 6489 W: 1245 L: 1334 D: 3910
sprt @ 15+0.05 th 1 Bonus for connected passed pawns
15-02-05 Roc LatentQAttack diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 13045 W: 2583 L: 2654 D: 7808
sprt @ 15+0.05 th 1 My last try on this idea, with last SPSA values. S(22,22) for Bishop and S(17.17) for Knight.
15-02-05 SC tuned_tempo diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 8077 W: 1601 L: 1686 D: 4790
sprt @ 15+0.05 th 1 Bugfix for tempo evaluation. More tempo value in endgames.
15-02-05 SC tuned_tempo diff 
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]

Total: 12197 W: 2416 L: 2489 D: 7292
sprt @ 15+0.05 th 1 More tempo value in middlegames, bugfix
15-02-05 jos no_split_after_null diff 
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]

Total: 16620 W: 2935 L: 2997 D: 10688
sprt @ 15+0.05 th 3 Don't split after null move. This is most likely not a good place to split. Most of the times it is rejected anyways due to min split depth. This should reduce a bit the split/search overhead. My guess is, this may be of more benefit with a higher number of threads, but start testing with 3 threads.
15-02-06 lbr krkm diff 
LLR: 4.36 (-2.94,2.94) [-3.50,0.50]

Total: 40050 W: 6382 L: 6249 D: 27419
sprt @ 15+0.05 th 1 are KRKm also useless?
15-02-06 lbr kpkp diff 
LLR: 3.04 (-2.94,2.94) [-3.50,0.50]

Total: 50933 W: 8035 L: 7995 D: 34903
sprt @ 15+0.05 th 1 is KPKP useless?
15-02-06 uri not_prune_high diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 17122 W: 3333 L: 3393 D: 10396
sprt @ 15+0.05 th 1 Usually do not change the search but avoid stupid pruning when the score is a losing score. fix the problem of stupid scores in small depth at r1q1nr1k/pp1b2b1/n2p2pp/2pP1p2/2B4B/3Q1N1P/PPP1NPP1/1R3RK1 b - - 0 12
15-02-06 jos matimb diff 
ELO: 0.22 +-2.1 (95%) LOS: 57.8%

Total: 40000 W: 8007 L: 7982 D: 24011
40000 @ 15+0.05 th 1 Check the new values after 3rd SPSA session.
15-02-06 n_p SPSAKingSafety3 diff 
45052/50000 iterations

99879/100000 games played
100000 @ 60+0.05 th 1 Another SPSA-session on king safety.
15-02-06 jos matimb diff 
LLR: -2.96 (-2.94,2.94) [0.00,4.00]

Total: 90570 W: 15049 L: 14915 D: 60606
sprt @ 60+0.05 th 1 LTC: values after 2nd SPSA session. (If these pass, I will test the final values of the 3rd session against the new master.)
15-02-06 sg pawn_attack_threat3 diff 
ELO: 1.34 +-3.0 (95%) LOS: 80.7%

Total: 20000 W: 3989 L: 3912 D: 12099
20000 @ 15+0.05 th 1 Allow queen as defender. The test of sn which allows all pieces as defenders passed STC, but struggles with LTC. So lets measure the effect for each piece type separatly.
15-02-06 sg pawn_attack_threat3 diff 
ELO: 0.28 +-3.0 (95%) LOS: 57.1%

Total: 20000 W: 4006 L: 3990 D: 12004
20000 @ 15+0.05 th 1 Allow rook as defender
15-02-06 sg pawn_attack_threat3 diff 
ELO: 0.12 +-3.0 (95%) LOS: 53.1%

Total: 20000 W: 3933 L: 3926 D: 12141
20000 @ 15+0.05 th 1 Allow bishop as defender
15-02-06 sg pawn_attack_threat3 diff 
ELO: 2.69 +-3.0 (95%) LOS: 95.9%

Total: 20000 W: 4072 L: 3917 D: 12011
20000 @ 15+0.05 th 1 Allow knight as defender
15-02-06 sg pawn_attack_threat3 diff 
ELO: 1.73 +-3.1 (95%) LOS: 86.2%

Total: 19246 W: 3921 L: 3825 D: 11500
20000 @ 15+0.05 th 1 Allow king as defender
15-02-07 lbr useless diff 
LLR: -2.96 (-2.94,2.94) [-3.50,0.50]

Total: 54033 W: 8354 L: 8630 D: 37049
sprt @ 15+0.05 th 1 is a combo of useless endgames still useless?
15-02-07 mco a7592e69d728ac839f098f2 diff 
LLR: 3.32 (-2.94,2.94) [-3.00,1.00]

Total: 146680 W: 23253 L: 23307 D: 100120
sprt @ 15+0.05 th 1 Verify that KQKRPs does not regress (8moves book to steer towards endgames)
15-02-07 vin passed_blockers2_spsa diff 
12234/12500 iterations

25000/25000 games played
25000 @ 15+0.05 th 1 Previous tuning run showed a trend in the rook, so try a second run expanded to include the minor pieces and the endgame. 25K games to find any trend (e.g. *should* the minor pieces be included or was my original suspicion that no bonus is needed correct)
15-02-07 vin pawns_both_wings diff 
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]

Total: 22251 W: 4488 L: 4533 D: 13230
sprt @ 15+0.05 th 1 Try an idea of Mindbreaker's - that the side ahead in an endgame (or approaching it) should try to preserve its pawns on both wings. The resulting eval is still symmetric. Two approaches seem possible - direct eval bonus or scaling factor adjust. Take 1 is the direct approach.
15-02-07 jos matimb diff 
ELO: -0.57 +-3.0 (95%) LOS: 35.6%

Total: 20000 W: 3984 L: 4017 D: 11999
20000 @ 15+0.05 th 1 Pit the final values of 3rd tuning session against those of the 2nd one.
15-02-07 sg pawn_attack_threat3 diff 
ELO: -0.81 +-2.6 (95%) LOS: 27.2%

Total: 27059 W: 5374 L: 5437 D: 16248
30000 @ 15+0.05 th 1 Allow knight, king and queen as defender. Combine the pieces which show some elo gain and measure if this adds up.
15-02-07 vin pawns_both_wings_alt diff 
LLR: -2.95 (-2.94,2.94) [-1.50,4.50]

Total: 16207 W: 3224 L: 3286 D: 9697
sprt @ 15+0.05 th 1 Try an idea of Mindbreaker's - that the side ahead in an endgame (or approaching it) should try to preserve its pawns on both wings. The resulting eval is still symmetric. Take 2 is the scale factor approach, which is cleaner, so it would be nice if this worked rather than take 1.
15-02-07 sg pawn_attack_threat3 diff 
LLR: -2.94 (-2.94,2.94) [-1.50,4.50]

Total: 28215 W: 5739 L: 5767 D: 16709
sprt @ 15+0.05 th 1 Allow knight as defender. Retest with SPRT to check for luck in first run
15-02-07 vin passed_blockers2_spsa diff 
39792/40000 iterations

81000/80000 games played
80000 @ 15+0.05 th 1 The trial run showed continued movement for rook and knight in particular, so go for a longer run with the previous values as the starting point. Upwards movement for the heavy piece endgame scores hints that this is still a factor there. Also switch to a much nicer table-driven way of expressing the weighting.
15-02-07 SC search_tempo_game_phase diff 
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]

Total: 19773 W: 3911 L: 3964 D: 11898
sprt @ 15+0.05 th 1 Game phase based tempo value (faster version), take 1. More tempo value for endgames.
15-02-07 SC search_tempo_game_phase diff 
LLR: -2.97 (-2.94,2.94) [-1.50,4.50]

Total: 17531 W: 3508 L: 3567 D: 10456
sprt @ 15+0.05 th 1 Game phase based tempo value (faster version), take 2. More tempo value for middlegame.
15-02-07 SC search_tempo_king_dista diff 
LLR: -2.96 (-2.94,2.94) [-1.50,4.50]

Total: 23776 W: 4839 L: 4880 D: 14057
sprt @ 15+0.05 th 1 King distance based tempo value. More tempo if kings are distant.