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

Same weight for black and white nonPawnCorrection history

Browse source 
Since we don't have color dependent parameters in NNUE eval, it also has
no sense IMO to have color dependent parameters in correction histories.

Ideally a fixed depth search on a single thread should be determistic,
so delivering the same result (move) if we just flip colors on the
board.

Patch replaces 2 parameters (122 and 185) with just one value 154 (= the avg of the two).

Passed STC-non regression
https://tests.stockfishchess.org/tests/view/6740a63286d5ee47d953f656
LLR: 2.95 (-2.94,2.94) <-1.75,0.25>
Total: 122336 W: 31499 L: 31372 D: 59465
Ptnml(0-2): 336, 14535, 31301, 14658, 338

Passed LTC-non regression
https://tests.stockfishchess.org/tests/view/67419bae86d5ee47d953f7b6
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 101400 W: 25870 L: 25731 D: 49799
Ptnml(0-2): 78, 11109, 28166, 11290, 57

closes https://github.com/official-stockfish/Stockfish/pull/5698

Bench: 1215483
This commit is contained in:
pb00067 2024-11-24 16:05:04 +01:00 • committed by Disservin
parent d5a36a3c92
commit 713000c517
1 changed files with 4 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/search.cpp
View file

@ -1435,6 +1435,7 @@ moves_loop: // When in check, search starts here
|| (bestValue > ss->staticEval && bestMove))) // positive correction & no fail low
{
const auto m = (ss - 1)->currentMove;
static const int nonPawnWeight = 154;
auto bonus = std::clamp(int(bestValue - ss->staticEval) * depth / 8,
-CORRECTION_HISTORY_LIMIT / 4, CORRECTION_HISTORY_LIMIT / 4);
@ -1443,9 +1444,9 @@ moves_loop: // When in check, search starts here
thisThread->majorPieceCorrectionHistory[us][major_piece_index(pos)] << bonus * 162 / 128;
thisThread->minorPieceCorrectionHistory[us][minor_piece_index(pos)] << bonus * 148 / 128;
thisThread->nonPawnCorrectionHistory[WHITE][us][non_pawn_index<WHITE>(pos)]
<< bonus * 122 / 128;
<< bonus * nonPawnWeight / 128;
thisThread->nonPawnCorrectionHistory[BLACK][us][non_pawn_index<BLACK>(pos)]
<< bonus * 185 / 128;
<< bonus * nonPawnWeight / 128;
if (m.is_ok())
(*(ss - 2)->continuationCorrectionHistory)[pos.piece_on(m.to_sq())][m.to_sq()] << bonus;