The last moves in the PV are usually quite weak. They don’t get a lot of playouts.
> -----Original Message----- > From: [email protected] [mailto:computer-go- > [email protected]] On Behalf Of Darren Cook > Sent: Wednesday, May 20, 2009 11:39 PM > To: computer-go > Subject: Re: [computer-go] Reflections on a disaster > > > Do you have any indication, which can be derived from the playouts, > > that a position might deserve an extra allotment of thinking time? > > I've a half-finished article, called "Consistent PV Enchancement". This > was inspired by looking at the prime variation information that Many > Faces gives out (check View|Show Lookahead). > > E.g. if the top move is 55% to white, black's reply is 45%, etc. But > nearer the bottom of the prime variation it has become 49% to white, 51% > to black, then something fishy is going on, and it needs more time to > investigate. > > The article is still unfinished as I started to feel it was just a minor > tweak for time usage, rather than the big jump in strength I'd hoped > for. Often it shows a bad move choice half way down the prime variation, > but the conclusion at the start of the prime variation was still correct. > > Darren > > P.S. This was the case in one position I looked at today: almost the > last move in the P.V. suddenly switched from about 40% on previous moves > to 85% to black. But with 200,000 playouts, so quite a solid estimate. I > played the P.V. out to that point and it was quite right - good for > black. This was caused by a white mistake 3 moves before. Once I chose a > better move there it went back to being good for white, so the estimate > at the start of the prime variation seemed valid. > > > -- > Darren Cook, Software Researcher/Developer > http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic > open source dictionary/semantic network) > http://dcook.org/work/ (About me and my work) > http://dcook.org/blogs.html (My blogs and articles) > _______________________________________________ > computer-go mailing list > [email protected] > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
