amchess wrote: ↑Fri Jul 31, 2020 8:14 pm
Hi, Peter.
Those options are used in the eval, but also in the search.
So, they change they engine's behavior.
I see, Andrea. Looking forward to test it.
Thanks again.
Thanks, Andrea!
NNUE- Eval is Hybrid now too, isn't it?
Latest Cfish has the fine option to choose between "Pure" and "Hybrid" NNUE. That's a big help for analysis, as far as I'm concerned.
Thanks, Andrea!
NNUE- Eval is Hybrid now too, isn't it?
Latest Cfish has the fine option to choose between "Pure" and "Hybrid" NNUE. That's a big help for analysis, as far as I'm concerned.
There is the option "Use NNUE" and the behavior is exactly the same as Stockfish.
Andrea
Thanks, Andrea!
NNUE- Eval is Hybrid now too, isn't it?
Latest Cfish has the fine option to choose between "Pure" and "Hybrid" NNUE. That's a big help for analysis, as far as I'm concerned.
There is the option "Use NNUE" and the behavior is exactly the same as Stockfish.
Andrea
I know, Andrea. SF has had an hybrid- eval patch at begin of August. Since then NNUE- eval is used only in more or less balanced positions decided by a treshold- boundary.
Positions like the one (much discussed here in context to that hybrid- patch) from the game nr.5 A0-SF8 2017 21.Bg5!
rn3r1k/pn1p1ppq/bpp4p/7P/4N1Q1/6P1/PP3PB1/R1B1R1K1 w - - 0 21
are'nt solved with hybrid- eval, with NNUE they are.
Jörg Oster made a fork for SF NNUE-always even because of that known issue (of course not for that single one position only, but it's one of them to see, if NNUE or hard coded evel is used, even if Use NNUE is checked).