#mon2y_rs — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #mon2y_rs, aggregated by home.social.
-
That's #mon2y_rs beating me at #NoThanks!
Took ages. Built an 'arena' program to help me test different amounts of iterations, best turn policies, exploration constants. Might be expanded later to different reward functions too.
Reason that it's turns were awful was that it turns out the √2 exploration constant that's usually used in #mcts just wasn't exploring the very random paths sufficiently. C = 3, iterations = 10000 is playing well. I was expecting 10-30x that would be needed.