Re: computer-go: Time management

[Dan Schmidt <dfan@xxxxxxxxxxxxxxxxx>]

Sai To Wang <swang@xxxxxxxxxxxxxxxxx> writes:

| I guess that for chess, after opponent play, about 90% of the tree
| will be discarded, and may be 99% for Go.
| 
| But still, the opponent time may be useful for Life&Death analisis.

In general, what computer chess programs do is to make an assumption
about the opponent's upcoming move, and expand the tree from there,
ignoring any other possibilities.

That way, if you guessed the opponent's move right, you can move
basically instantaneously, and if you guessed wrong, it's back to
square one.

This turns out to work better in the long run (over the course of a
game) than expanding the tree from the current position, taking all of
the opponent's moves into account, in which case you always save a
little bit of time.

I'm not sure how often a good go program would guess correctly as
compared to a good chess program.  The board is more open-ended in go,
but you have longer series of effectively forced moves at a time.

-- 
Dan Schmidt | http://www.dfan.org