So, after I replaced the endgame detection in QS with this I got about 8 Elo. It might be that the draw detection in QS was not necessary and costly (ThreeFoldRepetition was using ~5% of CPU on KPK positions).
I'm still debating whether I should or should not merge the change. IMO, 200 lines is too much for 8 Elo. I will run a gauntlet between stockfish and zurichess with and without kpk table as Evert suggested.
[1] https://bitbucket.org/brtzsnr/zurichess ... ew-default
kpk bitbase, worthless?
Moderator: Ras
-
- Posts: 433
- Joined: Fri Jan 16, 2015 4:02 pm
Re: kpk bitbase, worthless?
zurichess - http://www.zurichess.xyz