Re: On Game Space Size

[Avijit Ghosh <aghosh@xxxxxxxxxxxxxxxxx>]

On Thu, 29 Apr 1999, Avijit Ghosh wrote:

> or something like 8 million MC runs you have to do?? It would be
> nice to sample over that rather than just running all those
> configurations.. 

	Just to answer my own question, might it be possible
to sample over the variance? ie have a transition probability
from b,w configuration 1 -> b,w configuration 2 that
works to minimize the variance of those particular boards configurations
so you store a variance of each b,w configuration  you've
done your runs on which...  get minimized over time..and
you switch configurations using the variance as your test
criterea.. ie if you are sampling a configuration w/ a low
variance (ie converged) you have a high probability of
switching over to a configuration w/ a high variance and
over the course of a run you should uniformly converge all
possible configurations at once..  this still might be 
too expensive tho..
	I keep thinking you should sample only over configurations w/ a
high chance of valid configurations (ie some sort of importance sampling)
but am not sure how that would work for calculating the percentage 
valid boards / total boards if it does at all :)

	-avi