Magnus Persson wrote:
Quoting Matt Gokey <[EMAIL PROTECTED]>:
(snip)
Good point. This leads to another thought that I have been wondering
about. That is I question whether using more time to search more
simulations in the opening is the best approach. For the opening,
selecting reasonable robust moves that tend to lead to more favorable
options is probably a good objective. The lengths of the simulation
are perhaps too long to expect anything better. Later towards the
pre-middle to middle game it is very critical to play such that the
positions tactical potential is exploited such to secure connections
and eye space, etc. It would seem to me that focusing the highest
concentration of time and number of simulations during this part of
the game might be most advantageous.
It would be interesting for someone with a decent MC player to do an
experiment like this with one version concentrating highest number of
simulations in the opening and one concentrating in the middle game,
but otherwise equal and see which version wins more often.
My experience with Valkyria is that most time must be allocated towards
early moves. If you make a mistake on 9x9 it is very hard to come back. MC/UCT
programs such as MoGo are very good at defending a won position. Therefore it
is important to get a won position as early as possible. The longer it
thinks the better it plays. In the opening it always critical to find the best
move - but later on games are often either won or lost so saving time for those
positions are not so important.
I agree with this last statement for the early end-game / end-game
phase. For the pre-middle to middle I'm not sure. This is the
critical part of the game where you need to solidify the winning
positions. As long as reasonable opening moves that provide robust
options are made and it can play into the middle game stronger than the
opponent, winning positions should result.
I'm not saying not to spend time up front, just less than when in the
middle. On 9x9, the stage where this becomes tactically critical is
probably much earlier than on 19x19. Even with 9x9 I predict an MC/UCT
program that spent X simulations per move for the first 8 moves, and 2X
for the next 8 would tend to beat the same player doing the reverse,
where X is some reasonably large number of simulations needed to open
decently. I could be wrong, but it would be an interesting experiment
anyway. However IMO 9x9 is too small to see the real complications come
out.
Valkyria saves time in the opening by using an opening library, but as
soon as it leaves the library it spends a lot of time on the first move. Moves it
spends a lot of time on is also stored in the library. And later on I might
correct moves that was played in lost hand myself. I am actually rarely
satisfied with the opening moves of Valkyria. It needs more time for those
moves...
I was wondering if anyone was combining an opening library with MC/UCT
by running a large number of the simulations and storing the results.
This seems like a pretty natural extension to save simulation time in
the opening.
How strong a player are you? You are probably unfairly evaluating
Valkyria based on your strong expert play/perspective. I'm rather
amazed that MC simulations find good moves at all given that most of the
playouts are nonsense games. That is why I say MC/UCT is finding
reasonably robust moves with more favorable options (strategic play) not
necessarily great/best moves. Because of mostly meaningless playouts it
misses nuances and tactics deeper into the game that would show
otherwise. It seems the deeper the simulations the worse this effect
would be.
-Matt
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
