Param Tuning

Book: winrate 65_85
TC: 400+4
Total/Win/Draw/Lose: 24904 / 6257 / 12605 / 6042
PTNML: 2 / 2202 / 7834 / 2407 / 7
WinRate: 50.43%
ELO: 2.75[0.76, 4.71]
LOS: 99.64
LLR: 3.10[-2.94, 2.94]

Tuning done by afkbad

src/evaluate.cpp

@@ -49,14 +49,14 @@ Value Eval::evaluate(const Eval::NNUE::Network& network,
     int   nnueComplexity    = std::abs(psqt - positional);
 
     // Blend optimism and eval with nnue complexity
-    optimism += optimism * nnueComplexity / 556;
-    nnue -= nnue * nnueComplexity / 9359;
+    optimism += optimism * nnueComplexity / 481;
+    nnue -= nnue * nnueComplexity / 11222;
 
-    int mm = pos.major_material() / 43;
-    int v  = (nnue * (480 + mm) + optimism * (107 + mm)) / 483;
+    int mm = pos.major_material() / 42;
+    int v  = (nnue * (418 + mm) + optimism * (88 + mm)) / 518;
 
     // Damp down the evaluation linearly when shuffling
-    v -= (v * pos.rule60_count()) / 256;
+    v -= (v * pos.rule60_count()) / 250;
 
     // Guarantee evaluation does not hit the mate range
     v = std::clamp(v, VALUE_MATED_IN_MAX_PLY + 1, VALUE_MATE_IN_MAX_PLY - 1);