[Computer-go] quiescence in UCT search

Brian Sheppard sheppardco at aol.com
Tue Nov 8 17:43:12 PST 2011


That is what the UCB bound does. 

-----Original Message-----
From: computer-go-bounces at dvandva.org
[mailto:computer-go-bounces at dvandva.org] On Behalf Of Dave Dyer
Sent: Tuesday, November 08, 2011 8:11 PM
To: computer-go at dvandva.org
Subject: [Computer-go] quiescence in UCT search


My starting point is that the root node is different from all other nodes.
The purpose of any particular search is to select the next move.  Once a
particular child is far enough behind the leaders, it's effectively
eliminated, and any additional effort spent to investigate it is a waste.

For example, suppose we're going to search for 10 seconds, 5 seconds have
passed, the leading node has 10,000 visits, and some other node has 100.
It's mathematically impossible for the weak node to ever replace the strong
one.

I'm looking for a mathematical framework for making that kind of decision on
an ongoing basis.  Ideally, the top level nodes will be eliminated one by
one, as the probability that they would have eventually been the winning
choice falls below a chosen threshold.

_______________________________________________
Computer-go mailing list
Computer-go at dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go




More information about the Computer-go mailing list