Some performance measures could be developed using TB testing. And I know of one set of experiments with lc0 that used TB knowledge, to train lc0 to perfectly* fit and predict TB targets over positions not trained with. > I would think it worthwhile definitely (at least try). > I think you have a great idea, of using a bottom framework where we know the solutions for all legal positions and legal successor positions (=move). > I think all sorts of metrics comparisons could be done using TB as external absolute referential system. * about 99.97% on about a quarter of the 6 man positions tried, by said in #2: I don't expect the full game to simplify the gamut of behaviors that could be already considered interesting for engine testing in simpler TB chess. but starting small might be more intelligible, and likely surprises there, would have equivalent existence in the full game. Competitive measures could be tested for non-uniformity of play quality based on absolute performance measures. Great question, and you are not the only one thinking about it.Ī non competitive referential for testing. I always wondered why when toying with chess engines this opportunity had not much been explored. using human game databases, as I suspect you meant with real game.ĭistribution of testing positions could be characterized as departure from uniform. Perhaps that set of experiments could yield some candidates.Īlso, making scenarios of non-uniform position sampling for any type of engines, with given qualitative and/or quantitative characteristics, from e.g. I would think it worthwhile definitely (at least try). >The notion of real game though and uniformity, would have to be defined carefully. I think you have a great idea, of using a bottom framework where we know the solutions for all legal positions and legal successor positions (=move). I think all sorts of metrics comparisons could be done using TB as external absolute referential system.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |