deepcrayon
/
stockfish
1
0
Fork 0
Code Releases Activity

Smooth improving 

Smooth dependency on improvement margin in null move search.

STC
LLR: 2.93 (-2.94,2.94) <-0.50,2.50>
Total: 17384 W: 4468 L: 4272 D: 8644
Ptnml(0-2): 42, 1919, 4592, 2079, 60
https://tests.stockfishchess.org/tests/view/61689b8a1e5f6627cc1c0fdc

LTC
LLR: 2.94 (-2.94,2.94) <0.50,3.50>
Total: 45648 W: 11525 L: 11243 D: 22880
Ptnml(0-2): 26, 4731, 13036, 4997, 34
https://tests.stockfishchess.org/tests/view/6168a12c1e5f6627cc1c0fe3

It would be interesting to test if the other pruning/reduction heuristics
in master which are using the improving variable (ie the sign of improvement)
could benefit from a smooth function of the improvement value (or maybe a
Relu of the improvement value).

closes https://github.com/official-stockfish/Stockfish/pull/3740

Bench: 4916775
pull/3742/head
Stefano Cardanobile 2021-10-14 22:26:42 +02:00 committed by Stéphane Nicolet
parent 580698e5e5
commit 4231d99ab4
1 changed files with 12 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
src/search.cpp
View File

@ -590,7 +590,7 @@ namespace {
bool captureOrPromotion, doFullDepthSearch, moveCountPruning,
ttCapture, singularQuietLMR, noLMRExtension;
Piece movedPiece;
int moveCount, captureCount, quietCount, bestMoveCount;
int moveCount, captureCount, quietCount, bestMoveCount, improvement;
// Step 1. Initialize node
ss->inCheck = pos.checkers();
@ -766,6 +766,7 @@ namespace {
// Skip early pruning when in check
ss->staticEval = eval = VALUE_NONE;
improving = false;
improvement = 0;
goto moves_loop;
}
else if (ss->ttHit)
@ -804,13 +805,15 @@ namespace {
thisThread->mainHistory[~us][from_to((ss-1)->currentMove)] << bonus;
}
// Set up improving flag that is used in various pruning heuristics
// We define position as improving if static evaluation of position is better
// Than the previous static evaluation at our turn
// In case of us being in check at our previous move we look at move prior to it
improving = (ss-2)->staticEval == VALUE_NONE
? ss->staticEval > (ss-4)->staticEval || (ss-4)->staticEval == VALUE_NONE
: ss->staticEval > (ss-2)->staticEval;
// Set up the improvement variable, which is the difference between the current
// static evaluation and the previous static evaluation at our turn (if we were
// in check at our previous move we look at the move prior to it). The improvement
// margin and the improving flag are used in various pruning heuristics.
improvement = (ss-2)->staticEval != VALUE_NONE ? ss->staticEval - (ss-2)->staticEval
: (ss-4)->staticEval != VALUE_NONE ? ss->staticEval - (ss-4)->staticEval
: 200;
improving = improvement > 0;
// Step 7. Futility pruning: child node (~50 Elo).
// The depth condition is important for mate finding.
@ -826,7 +829,7 @@ namespace {
&& (ss-1)->statScore < 23767
&& eval >= beta
&& eval >= ss->staticEval
&& ss->staticEval >= beta - 20 * depth - 22 * improving + 168 * ss->ttPv + 177
&& ss->staticEval >= beta - 20 * depth - improvement / 15 + 168 * ss->ttPv + 177
&& !excludedMove
&& pos.non_pawn_material(us)
&& (ss->ply >= thisThread->nmpMinPly || us != thisThread->nmpColor))