[Computer-go] convergence in UCT search

Dave Dyer ddyer at real-me.net
Mon Nov 14 14:12:50 PST 2011


At 12:40 PM 11/14/2011, Brian Sheppard wrote:
>My understanding is the opposite: UCT search with random playouts converges
>to best play in every two-player game.

That is true with infinite time and memory.  In practice,
even very obviously non-optimum moves persist with any reasonable
amount of search resources.





More information about the Computer-go mailing list