5 ms·
How would you even go about making a model that can simulate a human chess skillset (saying levels implies that chess skillset is a scalar value while it is mor
by netdevnet 2y ago
How would you even go about making a model that can simulate a human chess skillset (saying levels implies that chess skillset is a scalar value while it is more reasonable to think of it as a tree of skills where your abilities might be higher or lower depend ending on the specific skill branch)
- anton-107 2y agotake millions of games of human players of certain rating only as your learning data?
- netdevnet 2y agoYou are assuming that's going to be a reliable proxy, what would make you think that?
- Someone 2y agoIn the context of this thread (“non-GM level computer chess”, which I read as also excluding International, FIDE Master, and Candidate Master (https://en.wikipedia.org/wiki/Grandmaster_(chess) https://en.wikipedia.org/wiki/Grandmaster_(chess))), I think it’s more important to not have a good learning algorithm. Even 10 thousand of such games may already have way more tactics than a player at the targeted level can detect and apply. If so, a learning algorithm that detects and remembers all of them already will be better than the target level.
- WithinReason 2y agocondition the move on ELO while training
- netdevnet 2y agoExactly. Level x (whatever scalar thing the user meant by that) doesn't quite work out for the reason you outlined. X Level Players have different tactics and someone that can use all of them will likely be better than most if not all those those players. I got downvoted for saying that. Maybe I didn't phrase it as well as you did
- wavemode 2y agoYeah but, won't it also be learning from the mistakes and missed tactics too? (Assuming its reward function is telling it to predict the human's move, rather than actually trying to win.)