On Dec 10, 2007 5:00 PM, Vlad Dumitrescu <[EMAIL PROTECTED]> wrote: > Am I confused in my understanding that a weakness of MC evaluation is > that due to its random play it will miss sequences where there is only > one winning move at each play? This is the way I am interpreting the > "nakade problem" discussed in another thread: to keep a dead-by-nakade > group dead, one must not miss one single move in the sequence. Of > course, in games the nakade example is just one of the simpler > variants, most semeai (capturing races) will fall into that category.
If a random playout played one side incorrectly and the other side correctly than you'd be correct that missing one move could have dire consequences. In random playouts, many strange combinations of moves can occur. Wins for both sides can occur. In random playouts, the side that has less ways to win will look like they're more likely to lose. This can lead to some significant bias in the probabilities from how an actual game would play out. There's ways to improve this, but nobody has made it perfect yet. One significant problem is that very smart playouts use up too much time and end up weakening the engine (in the general case). I have not followed this discussion very closely, but I think the argument is that one should focus on correcting the resulting MC probabilities rather than declaring the method incorrect. If the playouts don't have obviously stupid exchanges (and include all tesuji), they should be good estimators and be able to make extremely well educated decisions. Of course, that's not true in practice...
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
