Thanks,Lukasz.
DL -----Original Message----- From: Łukasz Lew <[EMAIL PROTECTED]> To: computer-go <[email protected]> Sent: Thu, 10 Jan 2008 7:47 am Subject: Re: [computer-go] On question about Libego110 The code of any version is easy to get: ttp://www.mimuw.edu.pl/~lew/hg/libego/?tags The file you are talking about is here: ttp://www.mimuw.edu.pl/~lew/hg/libego/?file/dfcd0a6db96e/uct.cpp f you take a look at line 151 you see: (bias should be renamed to umber_of_visits) explore_coeff = log (bias) * explore_rate; where bias is equivalent to this->bias i.e. number of visits in current node. lines lower is a loop over all children where we compute: float child_urgency = child->ucb <pl> (explore_coeff); where the body of ucb is: return (pl == player::black ? value : -value) + sqrt (explore_coeff / bias); and "bias" variable refers to the child's bias. If you have any more questions, just ask. ukasz On Jan 9, 2008 7:49 PM, <[EMAIL PROTECTED]> wrote: I'm looking at the code of Libgo110. I have a question. In the file uct.cpp and the definition of class note_t, the explore_coeff is calculated from log(node->bias). But in the paper 'Modification of UCT with Patterns in Monte-Carlo Go' table 1 line 10 -17, the explore_coeff is calculated from log(nb), where nb is the summation of node->bias for all the child nodes. Whyis the difference? Or did I read the code wrong? Thanks for any explanation. DL ________________________________ More new features than ever. Check out the new AOL Mail! _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ omputer-go mailing list [EMAIL PROTECTED] ttp://www.computer-go.org/mailman/listinfo/computer-go/ ________________________________________________________________________ More new features than ever. Check out the new AOL Mail ! - http://webmail.aol.com
_______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
