On 20-nov-08, at 14:03, Michael Williams wrote:
You could simply allow the pass all the time. Early in the game,
it will be a significantly inferior move and not often explored
in the tree. That may not be optimal, but it's certainly not
convoluted and you are guaranteed to never fail to generate the
pass move when you needed it.
I remember that's how I implemented it initially. I do not
remember the exact details of why I moved away from that, but I
believe that (in part) it had to do with the program occasionally
passing too early in the game.
When you say "too early", do you mean before the game is decided or
before all dead stones are removed. Only the first one is truly
too early.
In other words, there is no harm in leaving a dead stone on the
board given that it does not impact who the winner is.
I'm looking at my code and it's coming back to me. When deciding
which node to expand, which is a recursive process, I stop when
encountering two passes in a row. Instead of expanding the tree in
that case, I just determine the winner and update the win-visits
numbers in the tree. I believe I did this to allow for early
termination of the search. Without this, it would continue the search
until the time-allowance was depleted. This would cause a great waste
of thinking time towards the end of the game when the maximum
possible search-tree is much smaller than N.
But with the termination on two consecutive passes, what would
occasionally happen is that one side would pass and the simulation
would return a win. Then the other side could pass as well and based
on the number of stones still on the board, now the second player
would win. This caused problems. Therefore I decided only to allow a
pass towards the end of the game when it would also win when the
opponents stones still on the board are all considered alive.
I hope that makes sense :)
Mark
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
