Closed SFisGOD closed 3 years ago
the classical eval has a corned bishop term added explicitly (as the only FRC term, IIRC).
@SFisGOD See https://github.com/enedene/chess960-opening-book
@vondele It would be nice to upload an FRC book to https://github.com/official-stockfish/books
@SFisGOD not so easy, since fishtest can't play FRC, needs changes to the framework to instruct cutechess to play variants
This occurred in one of my FRC games. In the end Stockfish was essentially piece down and quickly lost. Its visually comparable to the snowball effect. By the way. The standard initial setup is one possible configuration within 960. So i would not call FRC a variant because Chess960 encompasses Chess.
@Nordlandia it needs -variant fischerandom
to be passed to cutechess (for move validation and whatever). So, right now it doesn't work out of the box. Also, consumers of fishtest games would need changes, since they don't expect variants. I would work out-of-the-box if we would remove all castling rights and just use the positions.
Has this issue has been recently addressed by a particular patch?
yes, at least practically, a term has been added to the NNUE eval that deals with this problem (gained about 40Elo for FRC). NNUE training is still not of on FRC games.
so, this is essentially fixed with the added term, will close the issue.
Stockfish lost in an FRC game against Allie https://tcec-chess.com/#div=frc2fl&game=1&season=19
The problem is the cornered bishop on a1. Stockfish classical has the CorneredBishop penalty to deal with this problem.
I browsed in https://github.com/official-stockfish/books and there is no FRC book. It would be nice to have an FRC book as well @noobpwnftw for FRC testing
Also, almost lost against Minic here https://tcec-chess.com/#div=frc2la&game=13&season=19 Stockfish would have lost this game against a stronger opponent.