[computer-go] Efficiently selecting a point to play in a random playout

[Don Dailey]

On Tue, 2007-06-05 at 15:58 -0400, Don Dailey wrote:
> You might improve the bias by "shuffling on the fly", perhaps when you
> find a legal move in the un-occupied point section of the list you
> could
> do a swap with the first move and a random move.   I'm wondering if
> the
> biased ordering of the list persists from move to move?   

With the presence of illegal moves, this is always biased even with some
shuffling (there is always some move more likely to be chosen that some
other move), but on the first pass (starting from a random list) it
doesn't matter, the bias itself is random (if that makes any sense.)
If the bias changes from move to move, then the whole thing is
effectively random.    But it's not time-effective to scramble the
un-occupied points section after every move.

Interesting stuff.  If avoiding RNG is a big saver, then I don't know if
my incremental shuffle slows it down too much.  I also am not sure
whether it introduces enough randomness to make a big difference
(because it still doesn't make it truly random, it just introduces some
chaos to helps the situation.)   

- Don