On Sun, Nov 16, 2008 at 11:46:28AM +0000, D Gilder wrote: > > This is the way I understand the random playouts: If, in a given position, > > white is clearly ahead, he will win the game if both parts play perfect > > moves. He is also likely to win if both parts play reasonably good moves > > (say, like human amateurs), but there is a bit more of a chance that one > > player hits upon a good combination which the other misses, so the result > > is not quite as reliable. If the playouts are totally random, there is > > still a better chance for white to win, because both parts make equally bad > > moves. The results have much more variation, of course. So far it does not > > sound like a very good proposal, but things change if you consider the > > facts that we don't have perfecr oracles, and good humans are slow to play > > out a position, and can not be integrated into a computer program. Whereas > > random playouts can be done awfully fast, tens of thousands of playouts in > > a second. Averaging the reuslts gives a fair indication of who is more > > likely to win from that position, just what is needed to decide which part > > of the search tree to expand. > > Do you know what use (if any) is made of the standard deviation of the > results?
Now statistics isn't my strong point, but one of the first and most successfull systems was called UCT for Upper Confidence Bound. It calculated some sort of expected error, and added that to the winning ratio. Then it expanded the branch that had the highest value. If that expansion was a win, the error margin would get smaller, but the average result would get higher. Perhaps some other branch would be tried next, but this one would still be a good candidate. If the result was a loss, the average would drop, and so would the error, so this move would become much less likely to be expanded. I am sure someone who understands statistics will soon jump in and correct my explanation :-) - Heikki -- Heikki Levanto "In Murphy We Turst" heikki (at) lsd (dot) dk _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
