search

amchess / ShashChess

A try to implement Alexander Shashin's theory on a Stockfish's derived chess engine
GNU General Public License v3.0
131 stars 28 forks source link

Question to shashchess author #49

Closed cj5716 closed 1 year ago

cj5716 commented 1 year ago

If a patch passes the 10-games 25 min + 10 second test and the test suite, however it fails the short time control SPRT in stockfish, will it be merged (considering shashchess author prioritises longer time controls?)

amchess commented 1 year ago

Clearly, completely random games, especially with nnue, imho, are useless. So, the priority is surely my strategy. Clearly, it must be more and more refined. In fact, we have testers creating smarter and smarter test suites, based on the win probability concept. This, too, is evidence of the limitations of the strategy of those at stockfish: the thousands of games are always the same, despite the fact that, with nnue, the vast majority of games end in draw. Unfortunately, top programmers are not top chess players and vice versa. It would take synergy, especially if the goal is not only to have a top engine, but also to be really useful to players of all levels and types. I am not at all interested in having a "who's got it longest" contest. By now, on the various engine rooms, even 4-year-old engines, with a good book, make it easy to pathetically....

amchess commented 1 year ago

If you need more infos and are interested in collaborating, you can contact me in private

cj5716 commented 1 year ago

Thank you, I was just interested in your testing methodology. Good luck!