Re: computer-go: What makes a ladder

["Njoroge" <Njoroge@xxxxxxxxxxx>]

> > I have been trying to gather rules about what makes a ladder and have
> > determined the following. (This can be in 1 of 8 orientations - rotated,
> > flipped and mirrored)
>
> I am afraid that in the end you need to read your ladder out. Kageyama
> starts his book Lessons in the Fundamentals of Go with ladders, pointing
out
> that there may be funny shortcuts that work many times, but learning them
> will only hurt your game, becasue reading them out is the only way that
> works every time, and learning to read is needed to improve your game
> anyway.

If a human reads out a ladder they can eventually determine whether a
capture can be made or not.  Following this rationale I see no reason why a
computer cannot do the same.  Reading ladders if far far simpler than
playing the game.  I think that given certain situations with breaking
stones the outcome of the ladder will be determine by running the isLadder
function of each of the breaker's liberties. If all liberties can be
captured by force then isLadder is true.

> So, a stone can be captured in a ladder if it is in atari, and no matter
how
> it struggles, can never get more than 2 liberties. Read it out. After all,
> the running part will always only have one move, and the attacking part
will
> only have two to choose from, one of them obviously wrong most of the
time.
> You need to check that the attacking stones do not end in atari
themselves,
> but even that ought to be simple, it can only happen near the last played
> move (of any color).

Using this diagram, when # plays at 'a' one of my rules states that O must
have at least 2 liberties.

 . L1.
 # a L2
 . O .


> Probably you should be willing to extend more advanced reading routines
with
> ladders where ever you meet them, ignoring any depth limitations...

Depth isn't a problem.  In the most complicated of ladders where stones can
be chased all the way round the board (given carful placing of other sontes)
still only recurses a maximum of say 50 or 60 times - well within time
constraints.