]> git.sesse.net Git - stockfish/commit
Tune search constants
authorxoto10 <buylow001@gmail.com>
Tue, 30 Jul 2019 10:46:43 +0000 (11:46 +0100)
committerStéphane Nicolet <cassio@free.fr>
Wed, 31 Jul 2019 07:35:20 +0000 (09:35 +0200)
commit8152a74ab4f703717fdb493cf9059f89be9a4fba
tree012c8a367b21e2026e400826ebbc6cd893c86b1d
parentd980d7c0d4b2efe7abe26bdd094859f6d888ee60
Tune search constants

This is the result of a 200k tuning run at LTC:
http://tests.stockfishchess.org/tests/view/5d3576b70ebc5925cf0e9e1e

which passed quickly at LTC:
LLR: 2.95 (-2.94,2.94) [0.50,4.50]
Total: 12954 W: 2280 L: 2074 D: 8600
http://tests.stockfishchess.org/tests/view/5d3ff3f70ebc5925cf0f87a2

STC failed, but second LTC at [0,4] passed easily:
LLR: 2.96 (-2.94,2.94) [0.00,4.00]
Total: 8004 W: 1432 L: 1252 D: 5320
http://tests.stockfishchess.org/tests/view/5d407cff0ebc5925cf0f9119

Further work?
No doubt some of these changes produce most of the gain and some are neutral
or even bad, so further testing on individual/groups of parameters changed
here might show more gains. It does look like these tests might need to be
at LTC though, so maybe not too practical to do. See the thread in the pull
request for an interesting discussion:
https://github.com/official-stockfish/Stockfish/pull/2260

Bench: 4024328
src/search.cpp