[computer-go] How to improve my minimax speed?

[David Doshay]

You can reduce this far more by using symmetry arguments:
if you are willing to eliminate only the edge moves from first
level considerations, then there are only 10 moves, six if you
are also willing to eliminate the second row, because of the
8-fold symmetry. While this advantage drops quickly as the
board fills with complicated patterns that break symmetry,
there are always the 8 boards that are instantly solved by the
info gained in doing the first ... but that is of little value if
there is no DB of previously searched patterns.

OOOOOOOOO
OOOOOOOOO
OOOOXOOOO
OOOXXOOOO
OOXXXOOOO
OXXXXOOOO
OOOOOOOOO

Cheers,
David



On 13, Nov 2006, at 1:29 AM, Eduardo Sabbatella wrote:

> This is the problem with Go. "Branching factor".
>
> 9x9=81,
> 81 x 80 x 79 = 511920 positions to check.
>
> You should try to search for "more likely" moves. Ex:
> A1 is definetely not a first move. Any move at first
> row/column will not be a good move, at least for the
> first 5 or 6 moves.
>
> So, just to ilustate:
>
> lets exclude the first row/column in the firsts moves.
>
> 7x7 = 49
> 49 x 48 x 47 = 110544 moves to check.
>
> That means: 79% improvement ONLY removing first
> row/column.
>