On Thu, Jan 20, 2011 at 8:50 AM, terry mcintyre <[email protected]>wrote:
> *From:* Stefan Kaitschick <[email protected]> > *Sent:* Thu, January 20, 2011 8:28:53 AM > * > * on 20.01.2011 05:35, Darren Cook wrote: > > Told you I was getting passionate about this point;-). Though actually > > I wasn't even thinking so much about number of playouts, more about > > other ideas such as first spending a few seconds on static semeai or > > life-death analysis on the root position then using that information to > > influence the playouts the same way RAVE values are used. > > > > Darren > > > Semeais and L&D are concerned with groups, eyes, libs, connections ect. > MCTS is concerned with winrates for moves. > > The gap between them is hard to bridge. At a minimum, the "why" will be > lost in translation. > > And if the "why" is lost, it becomes impossible to reason if the original > canditions are still met. > > This is especially significant because S+L&D information would be > propagated top-down, not botton-up like RAVE. > > I think the "next generation" bots will have to manage much more complex > information. It will be paradise lost. :-) > > > Stefan > > Perhaps the top-down analysis, not the bottom-up MCTS, should be in the > driver's seat, and the MCTS should be used to compare positions which are > considered "playable" according to the higher-level analysis? > It's going to be super difficult no matter how you slice it. Is there any way to have a pre-processor that looks at the starting board before the search and provides some direction that is passed to the tree search and playouts? Don > > > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
