kate/BadFish
1
0
Fork 0
mirror of https://github.com/sockspls/badfish synced 2025-04-29 16:23:09 +00:00
Code Issues Releases Wiki Activity

Introduction evaluation grain of 16 (and randomize)

Browse source 
This patch uses an evaluation grain of 16 in order to get more cutoffs in
the alpha-beta algorithm. For a discussion of the efficiency of alpha-beta
related to changes in the number of discrete values of terminal nodes, see
for instance section 9.1.2 of Judea Pearl's classical book "Heuristics" :

https://mat.uab.cat/~alseda/MasterOpt/Judea_Pearl-Heuristics_Intelligent_Search_Strategies_for_Computer_Problem_Solving.pdf

Moreover, we add a small (-1, +1) random component after the quantification
to help the search exploration a little bit. This is similar in spirit to
the (-1, +1) random component already present in the function draw_value()
to make Stockfish more robust in draw evaluations.

passed STC:
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 220960 W: 57249 L: 56668 D: 107043
Ptnml(0-2): 499, 26017, 56882, 26568, 514
https://tests.stockfishchess.org/tests/view/668907fb7edfb6f233f999f8

passed LTC :
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 48966 W: 12574 L: 12233 D: 24159
Ptnml(0-2): 14, 5233, 13654, 5562, 20
https://tests.stockfishchess.org/tests/view/6689105659cb3228a47598bf

closes https://github.com/official-stockfish/Stockfish/pull/5449

bench: 1336007
This commit is contained in:
Stéphane Nicolet 2024-07-06 12:16:38 +02:00 • committed by Joost VandeVondele
parent c40dd26cbc
commit d212e663bb
1 changed files with 3 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/evaluate.cpp
View file

@ -85,6 +85,9 @@ Value Eval::evaluate(const Eval::NNUE::Networks& networks,
int material = 554 * pos.count<PAWN>() + pos.non_pawn_material(); int material = 554 * pos.count<PAWN>() + pos.non_pawn_material();
v = (nnue * (73921 + material) + optimism * (8112 + material)) / 73260; v = (nnue * (73921 + material) + optimism * (8112 + material)) / 73260;
// Evaluation grain (to get more alpha-beta cuts) with randomization (for robustness)
v = (v / 16) * 16 - 1 + (pos.key() & 0x2);
// Damp down the evaluation linearly when shuffling // Damp down the evaluation linearly when shuffling
v -= v * pos.rule50_count() / 212; v -= v * pos.rule50_count() / 212;