| Finished | ||||||
|---|---|---|---|---|---|---|
| Swedis | Reckless | half_margin | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [-2.75, 0.25] Games: 25496 W: 6281 L: 6460 D: 12755 Ptnml(0-2): 77, 3077, 6591, 2954, 49 | perhaps we apply the adjustment in every node? |
| Swedis | Reckless | remove_incorrect_mat_draw | diff | 8.0+0.08 | LLR: 3.03 (-2.25, 2.89) [-2.75, 0.25] Games: 10954 W: 2772 L: 2623 D: 5559 Ptnml(0-2): 20, 1092, 3107, 1235, 23 | two knights is not always a draw, it can be mate in 1 |
| Swedis | Reckless | shallow_lmr_qs | diff | 8.0+0.08 | LLR: -2.43 (-2.25, 2.89) [0.00, 3.00] Games: 107098 W: 26717 L: 26625 D: 53756 Ptnml(0-2): 286, 12690, 27445, 12902, 226 | allow lmr to drop into qs for d2 searches |
| Swedis | Reckless | fp_corrplexity | diff | 8.0+0.08 | LLR: -2.27 (-2.25, 2.89) [0.00, 3.00] Games: 21172 W: 5187 L: 5275 D: 10710 Ptnml(0-2): 44, 2626, 5345, 2516, 55 | fp less in high corrplexity positions. base margin was adjusted to roughly compensate so the average margin stays roughly the same |
| Swedis | Reckless | truecumber_halfmove_corr | diff | 8.0+0.08 | LLR: -2.28 (-2.25, 2.89) [0.00, 3.00] Games: 10308 W: 1794 L: 1859 D: 6655 Ptnml(0-2): 3, 466, 4284, 395, 6 | idea from truecumber in TCEC chat, use halfmove hash for corrhist indexing |
| Swedis | Reckless | truecumber_halfmove_corr | diff | 8.0+0.08 | LLR: -0.57 (-2.25, 2.89) [0.00, 3.00] Games: 105974 W: 26296 L: 26103 D: 53575 Ptnml(0-2): 238, 11833, 28653, 12024, 239 | idea from truecumber in TCEC chat, use halfmove hash for corrhist indexing. BUGGED: regresses hash() method on board |
| Swedis | Reckless | hist_malus_quad | diff | 40.0+0.40 | LLR: -2.32 (-2.25, 2.89) [0.00, 3.00] Games: 87476 W: 21092 L: 21040 D: 45344 Ptnml(0-2): 23, 9911, 23822, 9955, 27 | scale down malus with moves^2 |
| Swedis | Reckless | draw_63 | diff | 8.0+0.08 | LLR: 0.08 (-2.25, 2.89) [0.00, 3.00] Games: 25754 W: 6446 L: 6384 D: 12924 Ptnml(0-2): 92, 3013, 6600, 3085, 87 | make draw scores fall on [-3,-2,-1,1,2,3] — needs rebasing |
| Swedis | Reckless | undefended_movepick_threats | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 3.00] Games: 32476 W: 8039 L: 8101 D: 16336 Ptnml(0-2): 84, 3922, 8284, 3868, 80 | dont give a bonus unless we're actually moving out of a threat |
| Swedis | Reckless | undefended_movepick_threats | diff | 8.0+0.08 | LLR: -2.56 (-2.25, 2.89) [0.00, 3.00] Games: 1234 W: 231 L: 392 D: 611 Ptnml(0-2): 11, 234, 278, 93, 1 | give the same bonus to moving undefended pieces out of danger as we give to pieces threatened by a less valuable piece. uses the setwise slider attacks by 87flowers on avx512 targets. broken! forgot to make sure that for a piece being undefended to matter it should also be threatened, otherwise we are giving bonuses to moves that are not in fact moving out of danger |
| Swedis | Reckless | hist_malus_quad | diff | 8.0+0.08 | LLR: 2.91 (-2.25, 2.89) [0.00, 3.00] Games: 60178 W: 15154 L: 14849 D: 30175 Ptnml(0-2): 154, 7069, 15360, 7330, 176 | scale down malus with moves^2 |
| Swedis | Reckless | hist_malus_log | diff | 8.0+0.08 | LLR: -2.32 (-2.25, 2.89) [0.00, 3.00] Games: 12132 W: 2924 L: 3033 D: 6175 Ptnml(0-2): 31, 1468, 3175, 1363, 29 | scale history move count malus with log(moves) |
| Swedis | Reckless | tt_hist_cutnode | diff | 8.0+0.08 | LLR: -2.29 (-2.25, 2.89) [0.00, 3.00] Games: 28030 W: 6977 L: 7051 D: 14002 Ptnml(0-2): 92, 3350, 7193, 3300, 80 | only update history for tt cutoffs in cutnodes |
| Swedis | Reckless | more_fds_reduction | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 3.00] Games: 1454 W: 300 L: 430 D: 724 Ptnml(0-2): 5, 237, 369, 115, 1 | simpler more aggro full depth search reduction |
| Swedis | Reckless | quadratic_fp | diff | 8.0+0.08 | LLR: -2.36 (-2.25, 2.89) [0.00, 3.00] Games: 42646 W: 10660 L: 10706 D: 21280 Ptnml(0-2): 151, 5130, 10782, 5134, 126 | add a quadratic term to fp, so fp is more aggro for low depths and less aggro for deeper depths |
| Swedis | Reckless | lmr_cutoffcnt_smooth | diff | 8.0+0.08 | LLR: -2.28 (-2.25, 2.89) [0.00, 3.00] Games: 7202 W: 1737 L: 1855 D: 3610 Ptnml(0-2): 22, 907, 1852, 807, 13 | reduce more with higher cutoff count |
| Swedis | Reckless | u8_nnz | diff | 4.0+0.04 | LLR: -2.29 (-2.25, 2.89) [0.00, 3.00] Games: 19270 W: 4944 L: 5035 D: 9291 Ptnml(0-2): 79, 2127, 5315, 2034, 80 | faster NNZ using the fact that L1 < 4 * 2^8 so the indices can be 8 bits instead of 16 my idea but impl is almost copied verbatim from the one anematode wrote https://tests.stockfishchess.org/tests/view/69855f88c5ad5e83c2ca2871 |