On 12:53 Thu 15 Jan     , Mark Boon wrote:
> 
> On Jan 15, 2009, at 12:33 PM, Daniel Waeber wrote:
> 
> > yes, but the weight/color maps stay the same for all updated nodes.
> >
> > I think the first playoutNode (the one most deep inside the tree) only
> > should get amaf values for the random playout, the next one form  
> > random
> > playout + from the first playoutNode ... and the root node amaf values
> > form all the nodes.
> 
> OK, I think I see now what you're trying to say. This is something I  
> did think about. I hope my memory serves me well enough to say why I  
> didn't do it that way.
> 
> - What you propose adds complexity and possibly computation (if it  
> means recalculating or adjusting the weight map).
> - I don't think it makes all that much difference.
> 
> The reason I don't think it makes much difference is that adding AMAF  
> values for the moves above (closer to the root) the playoutNode is  
> that most likely those points are now occupied. Since the AMAF values  
> are used to compute which empty points are the best next candidate,  
> the AMAF values at occupied points are immaterial. They are not even  
> added. So it only makes a difference in cases where played stones get  
> captured and those points then are occupied again. This brings us back  
> to the issue discussed earlier about ko and ishi-no-shita.

Yes, I don't know if it would be a big improvement. It is not a real
speed decrease, as you just move the point were you add the additional
values to the maps.
Perhaps I'll run some tests when I have the time.

> Mark
> 
> P.S. what do I need to open that file? Is it a SVN patch?

it's standard diff format, outputed by git. You can view it with a text
editor and apply it with unix "patch" commandline tool or use
eclipse->team->apply-patch. it's done agains svn-root, so you have to ignore
2 leading pathnames if you apply it inside the refbot folder.

regards
  wabu

_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to