1
0
Fork 0
mirror of https://github.com/sockspls/badfish synced 2025-04-29 16:23:09 +00:00

Separate eval params for smallnet and main net

Values found with spsa around 80% of 120k games at 60+0.6:
https://tests.stockfishchess.org/tests/view/669205dac6827afcdcee3ea4

Passed STC:
https://tests.stockfishchess.org/tests/view/6692928b4ff211be9d4e98a9
LLR: 2.96 (-2.94,2.94) <0.00,2.00>
Total: 313696 W: 81107 L: 80382 D: 152207
Ptnml(0-2): 934, 36942, 80363, 37683, 926

Passed LTC:
https://tests.stockfishchess.org/tests/view/6692aab54ff211be9d4e9915
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 228420 W: 57903 L: 57190 D: 113327
Ptnml(0-2): 131, 25003, 63243, 25688, 145

closes https://github.com/official-stockfish/Stockfish/pull/5486

bench 1319322
This commit is contained in:
Linmiao Xu 2024-07-13 00:40:07 -04:00 committed by Joost VandeVondele
parent de2bf1a186
commit e443b2459e

View file

@ -79,11 +79,11 @@ Value Eval::evaluate(const Eval::NNUE::Networks& networks,
}
// Blend optimism and eval with nnue complexity
optimism += optimism * nnueComplexity / 457;
nnue -= nnue * nnueComplexity / 19157;
optimism += optimism * nnueComplexity / (smallNet ? 433 : 453);
nnue -= nnue * nnueComplexity / (smallNet ? 18815 : 17864);
int material = 554 * pos.count<PAWN>() + pos.non_pawn_material();
v = (nnue * (73921 + material) + optimism * (8112 + material)) / 73260;
int material = (smallNet ? 553 : 532) * pos.count<PAWN>() + pos.non_pawn_material();
v = (nnue * (73921 + material) + optimism * (8112 + material)) / (smallNet ? 68104 : 74715);
// Evaluation grain (to get more alpha-beta cuts) with randomization (for robustness)
v = (v / 16) * 16 - 1 + (pos.key() & 0x2);