Sorry: each trial = tree-search + playout.
 
Within the tree Pebbles does a progressive bias calculation for many nodes,
but that only applies to a fraction of trials (about 40%, IIRC, and I
suspect that it could be reduced, but that doesn't matter much because
progressive bias is a low percentage of Pebbles).
 
Then RAVE applies to several nodes in each trial. But not all nodes. If the
node won on the last trial, then the best move is the same as before.
 
So the tree part has three pieces that are proportional to board size: RAVE
updating, progressive bias, and UCT+RAVE selection.


  _____  

From: [email protected]
[mailto:[email protected]] On Behalf Of Peter Drake
Sent: Wednesday, June 29, 2011 7:13 PM
To: [email protected]
Subject: Re: [Computer-go] A Linear Classifier Outperforms UCT on 9x9 Go


Wait ... how do you find the move with the highest UCT value (or whatever
you're using) without at least examining all of them? 

On Jun 29, 2011, at 2:39 PM, Brian Sheppard wrote:


A typical Pebbles playout does *no* work that is proportional to the number
of squares on the board


Peter Drake
http://www.lclark.edu/~drake/



_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to