In our SlugGo DB effort we have the DB return all moves known to
continue from the board state, and each move is also associated
with its winning percentage.
Cheers,
David
On 14, Oct 2006, at 5:31 PM, Don Dailey wrote:
There is another technique that may be more effective that the one I
have been considering for building a hybrid search/database solver.
Suppose we used the same basic idea of building a function that can be
given the veto by a bloom filter, but the function in question returns
a best MOVE instead of an evaluation function.
My guess is that it's easier to return a best move than it is to
return a correct and tightly bounded score. If this is so, we can get
by with a much smaller bloom filter.
But what do you do if the filter says you don't know what the best
move is? That's easy.
You start by building an evaluation function
that walks the path of the best moves until it reaches the terminal
position of the game. Now you can use this evaluation function to
find the best move in any unknown position by doing a 1 ply search,
using it as the terminal node evaluator! This evaluation
function may
recursively have to call itself whenever it encounters positions where
the best move is not known.
This technique demonstrates that there is little difference in knowing
the value of a position, or knowing the best move - they are
essentially
equivalent because one can be easily deduced (or converted) to the
other.
- Don
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
