| Finished | ||||||
|---|---|---|---|---|---|---|
| World | Reckless | razoring-improving-3 | diff | 8.0+0.08 | LLR: -1.93 (-2.25, 2.89) [0.00, 4.00] Games: 4546 W: 1050 L: 1126 D: 2370 Ptnml(0-2): 28, 593, 1095, 541, 16 | clamp(improvement, -512, 512) / 3 |
| World | Reckless | nmp_r_corrplex4 | diff | 8.0+0.08 | LLR: -1.73 (-2.25, 2.89) [0.00, 4.00] Games: 2172 W: 481 L: 555 D: 1136 Ptnml(0-2): 12, 304, 524, 238, 8 | 5 - correction_value / 150 after rebase |
| World | Reckless | razoring-improving-2 | diff | 8.0+0.08 | LLR: -1.16 (-2.25, 2.89) [0.00, 4.00] Games: 5898 W: 1382 L: 1417 D: 3099 Ptnml(0-2): 23, 739, 1464, 696, 27 | 250 * depth * improving |
| World | Reckless | pcm-prior-reduction | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 45682 W: 10695 L: 10660 D: 24327 Ptnml(0-2): 178, 5505, 11453, 5514, 191 | as Pere mentioned the depth is reduced in the current ply, so undo it proportionally to the prior reduction |
| World | Reckless | two-killers | diff | 8.0+0.08 | LLR: -2.25 (-2.25, 2.89) [0.00, 4.00] Games: 14214 W: 3322 L: 3382 D: 7510 Ptnml(0-2): 62, 1767, 3502, 1721, 55 | |
| World | Reckless | more-base-reduction | diff | 8.0+0.08 | LLR: -2.25 (-2.25, 2.89) [0.00, 4.00] Games: 17298 W: 4076 L: 4127 D: 9095 Ptnml(0-2): 88, 2119, 4278, 2084, 80 | increase base LMR reduction by 200 (1/5th depth) |
| World | Reckless | new-depth-thing | diff | 8.0+0.08 | LLR: -2.25 (-2.25, 2.89) [0.00, 4.00] Games: 25446 W: 6040 L: 6066 D: 13340 Ptnml(0-2): 120, 3070, 6363, 3056, 114 | |
| World | Reckless | simplify-nmp-formula | diff | 8.0+0.08 | LLR: 2.90 (-2.25, 2.89) [-4.50, 0.00] Games: 11966 W: 2889 L: 2815 D: 6262 Ptnml(0-2): 50, 1291, 3231, 1357, 54 | simplify NMP formula regarding TT move |
| World | Reckless | L1-640 | diff | 40.0+0.40 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 6692 W: 1569 L: 1649 D: 3474 Ptnml(0-2): 16, 845, 1696, 781, 8 | |
| World | Reckless | lmr-cn-tt-move | diff | 8.0+0.08 | LLR: -2.40 (-2.25, 2.89) [0.00, 4.00] Games: 5672 W: 1311 L: 1405 D: 2956 Ptnml(0-2): 27, 747, 1382, 653, 27 | |
| World | Reckless | lmr-neg-ext | diff | 8.0+0.08 | LLR: -2.33 (-2.25, 2.89) [0.00, 4.00] Games: 5088 W: 1147 L: 1238 D: 2703 Ptnml(0-2): 29, 645, 1276, 576, 18 | naively use se results in lmr |
| World | Reckless | eval-scale | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 6530 W: 1505 L: 1589 D: 3436 Ptnml(0-2): 29, 840, 1611, 756, 29 | use 90% of eval |
| World | Reckless | rfp-tt-quiet | diff | 8.0+0.08 | LLR: -1.48 (-2.25, 2.89) [0.00, 4.00] Games: 3142 W: 703 L: 763 D: 1676 Ptnml(0-2): 25, 411, 744, 381, 10 | guard RFP when tt move is quiet |
| World | Reckless | L1-640 | diff | 8.0+0.08 | LLR: -1.71 (-2.25, 2.89) [0.00, 4.00] Games: 2276 W: 522 L: 593 D: 1161 Ptnml(0-2): 9, 303, 589, 224, 13 | train NNUE with 128 more neurons in L1 |
| World | Reckless | L1-640 | diff | N=25000 | LLR: 2.91 (-2.25, 2.89) [0.00, 4.00] Games: 4374 W: 1474 L: 1274 D: 1626 Ptnml(0-2): 135, 455, 841, 587, 169 | train NNUE with 128 more neurons in L1 |
| World | Reckless | rfp-tt-quiet | diff | 8.0+0.08 | LLR: -1.53 (-2.25, 2.89) [0.00, 4.00] Games: 4530 W: 1034 L: 1091 D: 2405 Ptnml(0-2): 22, 585, 1104, 536, 18 | less RFP when tt move is quiet |
| World | Reckless | stat-score-pcm | diff | 8.0+0.08 | LLR: -2.29 (-2.25, 2.89) [0.00, 4.00] Games: 25258 W: 5904 L: 5932 D: 13422 Ptnml(0-2): 110, 3063, 6304, 3049, 103 | adjust pcm factor based on move's history |
| World | Reckless | nmp-base-depth | diff | 8.0+0.08 | LLR: -2.52 (-2.25, 2.89) [0.00, 4.00] Games: 15324 W: 3534 L: 3602 D: 8188 Ptnml(0-2): 64, 1867, 3863, 1809, 59 | increase base depth 4 -> 5 |
| World | Reckless | cutoff-failures-lmr | diff | 8.0+0.08 | LLR: -2.30 (-2.25, 2.89) [0.00, 4.00] Games: 12466 W: 2823 L: 2890 D: 6753 Ptnml(0-2): 47, 1546, 3120, 1467, 53 | reduce cutnode reduction in proportion to the number of moves played (failures to cause a beta cutoff) |
| World | Reckless | guard-cn-lmr | diff | 8.0+0.08 | LLR: -2.11 (-2.25, 2.89) [0.00, 4.00] Games: 12478 W: 2904 L: 2963 D: 6611 Ptnml(0-2): 51, 1566, 3076, 1483, 63 | guard the cutnode reduction if nmp/probcut failed |
| World | Reckless | rfp-failfirm | diff | 8.0+0.08 | LLR: -2.45 (-2.25, 2.89) [0.00, 4.00] Games: 10994 W: 2546 L: 2625 D: 5823 Ptnml(0-2): 48, 1384, 2705, 1319, 41 | beta + (eval - beta) / 3 |
| World | Reckless | nmp-static-eval | diff | 8.0+0.08 | LLR: -2.27 (-2.25, 2.89) [0.00, 4.00] Games: 63218 W: 14853 L: 14766 D: 33599 Ptnml(0-2): 249, 7595, 15838, 7674, 253 | use staticeval against the common sense |
| World | Reckless | lmp-prior-reduction | diff | 8.0+0.08 | LLR: -2.25 (-2.25, 2.89) [0.00, 4.00] Games: 6530 W: 1517 L: 1601 D: 3412 Ptnml(0-2): 20, 865, 1594, 751, 35 | search less when the node has been reduced more, as it's probably not as important |
| World | Reckless | early-reduction-history | diff | 8.0+0.08 | LLR: -2.27 (-2.25, 2.89) [0.00, 4.00] Games: 50370 W: 11733 L: 11684 D: 26953 Ptnml(0-2): 211, 6103, 12522, 6124, 225 | another attempt to move reduction adjustments before pruning |
| World | Reckless | fix-evasion-scoring | diff | 8.0+0.08 | LLR: 2.93 (-2.25, 2.89) [-5.00, 0.00] Games: 15214 W: 3552 L: 3505 D: 8157 Ptnml(0-2): 51, 1637, 4209, 1634, 76 | there was a hack to use the pawn value as the captured piece in noisy move scoring (since the ep square is empty). but this breaks promotions (their empty square is actually empty), so fix it |