Hi,
Well maybe Sylvain is willing to answer some related
questions of mine. Mogo seems to be much stronger then
the rest, so I am more then curious what is more important:
1) A clever "random" playaout (I assume this is called
a proper distribution now?)
OR
2) Clever things within the UTC tree and/or above that.
(Go knowlegde, selectivity etc.)
As long as you have a proper tree search algorithm as UCT with a good
parameter constant, the distribution of the simulations (so the
simulation player, or simulation policy) is much more important. I
truly believe that we can make big steps in this direction and getting
much better (real not simulation) players that way.
And: is the answer the same for 9x9 and 19x19 boards?
In the history of MoGo, every improvement in 9x9 was an improvement in
19x19, and there are very few specifics things for 19x19 (as playing
local).
I believe that improving the simulation policy can (will ?) bring
impressive improvements in 9x9 and even more in 19x19.
I am very optimistic for that.
Yet, improving UCT (or other tree search algorithm) is clearly
possible, and certainly useful once you have a "very" good simulation
policy.
Sylvain
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
