wat
New Member
Posts: 32
|
Post by wat on Mar 8, 2024 19:29:33 GMT -8
Hey, I doubt this thread will come back to life, but I have to ask, after reading all this - what "Subtle flaw" was corrected that fixed the problem? The link referenced is dead, so there's no hint of what the issue was. Quite disappointing, reaching the end with no answer, as op seems to describe what I'm facing. "Eventually I realised that (in effect) the player chooses a move on the assumption that the opponent will play randomly, and MCTS provides no means of avoiding moves that result in an immediate loss." The assumption that opponents play randomly is inaccurate in most cases. I guess dpoly figured that out at some point and fixed the implementation. About the thread coming back to life, it's a pity GGP servers are all down nowadays.
|
|
|
Post by alandau on Mar 8, 2024 20:38:57 GMT -8
Hey, I doubt this thread will come back to life, but I have to ask, after reading all this - what "Subtle flaw" was corrected that fixed the problem? The link referenced is dead, so there's no hint of what the issue was. Quite disappointing, reaching the end with no answer, as op seems to describe what I'm facing. I don't know what flaw he's talking about in particular, but I believe the paper that was linked to was "A Survey of Monte Carlo Tree Search Methods" by Cameron Browne. There seem to be many copies online, e.g.: www.lamsade.dauphine.fr/~cazenave/A+Survey+of+Monte+Carlo+Tree+Search+Methods.pdf
|
|