Tags: xu-shawn/Serendipity
Tags
Fix Illegal PV (#115) Elo | 0.98 +- 2.90 (95%) SPRT | 8.0+0.08s Threads=1 Hash=8MB LLR | 2.96 (-2.94, 2.94) [-5.00, 0.00] Games | N: 16686 W: 4079 L: 4032 D: 8575 Penta | [169, 1830, 4300, 1873, 171] https://chess.aronpetkovski.com/test/2832/ bench 1849271
Update WDL Model (#92) 6x More data from various non-regression tests mainly relating to resizable hash. https://chess.aronpetkovski.com/test/1225/ https://chess.aronpetkovski.com/test/1250/ https://chess.aronpetkovski.com/test/1261/ https://chess.aronpetkovski.com/test/1263/ https://chess.aronpetkovski.com/test/1277/ https://chess.aronpetkovski.com/test/1357/  ``` ❯ python3 scoreWDL.py --NormalizeData '{"momType": "material", "momMin": 17, "momMax": 78, "momTarget": 58, "as": [-40.00676365, 98.77222217, -182.08209988, 340.61197659]}' Converting evals with NormalizeData = {'momType': 'material', 'momMin': 17, 'momMax': 78, 'momTarget': 58, 'as': [-40.00676365, 98.77222217, -182.08209988, 340.61197659]}. Reading eval stats from scoreWDLstat.json. Retained (W,D,L) = (824989, 1928200, 825371) positions. Fit WDL model based on material. Initial objective function: 0.6350481731497092 Final objective function: 0.6350379750816926 Optimization terminated successfully. const int NormalizeToPawnValue = 236; Corresponding spread = 132; Corresponding normalized spread = 0.558809843707221; Draw rate at 0.0 eval at material 58 = 0.7137362232663982; Parameters in internal value units: p_a = ((-117.298 * x / 58 + 388.530) * x / 58 + -512.962) * x / 58 + 477.991 p_b = ((-47.751 * x / 58 + 118.817) * x / 58 + -33.058) * x / 58 + 94.017 constexpr double as[] = {-117.29796837, 388.52982654, -512.96172771, 477.99147478}; constexpr double bs[] = {-47.75093113, 118.81734426, -33.05800930, 94.01690687}; ``` bench 2164850