[computer-go] More UCT / Monte-Carlo questions

[Jason House]

On Feb 5, 2008 1:15 PM, Olivier Teytaud <Olivier.Teytaud at lri.fr> wrote:

> > I think mogo is the only strong program that uses the UCB1-tuned
> formula.
> > The others use the same formula you use.  I found a thesis where they
> > measured many different formulas and found little difference.   If any
> > strong program other than mogo uses some formula other than the basic
> one,
> > can you please let us know?
> >
>
> MoGo does not use the UCB1 formula or the UCB1-Tuned formula
> (see ICML paper from Gelly and Silver). Also, now, the formula
> has been tuned again and again and does not look anymore like
> something I could explain :-) this is almost genetic programming :-)
>
> Basically, the formula in MoGo combines the success ratio and the
> RAVE-success ratio, with more focus on the success ratio when the
> number of simulations is large.


Please correct me if I'm wrong, but I thought:
  Only once you combine the two components does the influence of UCB1 and
UCB1-tuned become less obvious.  If you look at just the RAVE success ratio
component, or just the success ratio component, I believe the UCB1-Tuned
formula is still present.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20080205/28982cc8/attachment.htm