]> git.sesse.net Git - stockfish/commit
Tweak kingAttackersCount and KingAttackWeights
authorStéphane Nicolet <cassio@free.fr>
Sun, 13 May 2018 05:20:39 +0000 (07:20 +0200)
committerStéphane Nicolet <cassio@free.fr>
Sun, 13 May 2018 05:20:39 +0000 (07:20 +0200)
commitbf01bd07551dd4e2d68d82b305cc5d89a4c23bb0
treea4d25265b1ea99cb74b544b0585ba82ff59e839c
parentc163c2fcb0bf6fbb2c41b1399e7928cb76cd6118
Tweak kingAttackersCount and KingAttackWeights

Use the whole kingRing for pawn attackers instead of only the squares directly
around the king. This tends to give quite a lot more kingAttackersCount, so to
compensate and to avoid raising the  king danger too fast we lower the values
in the KingAttackWeights array a little bit.

STC:
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 51892 W: 10723 L: 10369 D: 30800
http://tests.stockfishchess.org/tests/view/5af6d4dd0ebc5968e652428e

LTC:
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 24536 W: 3737 L: 3515 D: 17284
http://tests.stockfishchess.org/tests/view/5af709890ebc5968e65242ac

Credits to user @xoroshiro for the idea of using the kingRing for pawn attackers.

How to continue? It seems that the KingAttackWeights[] array stores values
which are quite Elo-sensitive, yet they have not been tuned with SPSA recently.
There might be easy Elo points to get there.

Closes https://github.com/official-stockfish/Stockfish/pull/1597

Bench: 5282815
AUTHORS
src/evaluate.cpp