Re: [gnugo-devel] safety_3_10.1

[Arend Bayer]

Dan wrote:

> > I like the idea, but have some doubts whether it can be made work well.
> > I found at least 3 kind of situations where it would currently produce
> > non-sense. I've appended them below. Also, I generally have some
> > distrust towards valuations that cannot possibly be made continuous...
>
> The condition of having 2 eyes cannot be made continuous!
>
> I think the surround status is a potentially useful
> measure, even though it is inherently discrete. There is a
> *qualitative* difference between the situation where a
> dragon is enclosed by the convex hull of its neighbors,
> and when it is not. If it is not enclosed by the convex
> hull of its neighbors, it is difficult to contain it
> with a single move for purely geometrical reasons.

Well, I didn't mean to say it won't work, just that it is difficult to
make it work well. I think a very good example in usual play are
reduction moves of moyos. These are almost always played just at the
boundary of the convex hull of the stones surrounding the moyo.

I think the hard part is to select the relevant hostile stones.

> A picture is worth 1000 words. In nngs:8, the dragon at M7 is
> surrounded by this criterion. Here is the expanded hull:
(You mean nngs1:8):

   A B C D E F G H J K L M N O P Q R S T
19 . . . . . . . . . . . . . . . . . . . 19
18 . . . . . . . . . . . . . . . . . . . 18
17 . . . . . . . . . . . . X . . . . . . 17
16 . . . O . . . . . + . . . . . X . . . 16
15 . . . . . . . . . . . . . . . . . . . 15
14 . . . . . . . . . . . . . . . . X . . 14
13 . . . . . . . . . . . . . . . O O . . 13
12 . . . . . . . . . . . . . . . . . . . 12
11 . . . . . . . . . . . . . . . . . . . 11     WHITE has captured 0 stones
10 . . O + . . . . . + . . . . . + . . . 10     BLACK has captured 0 stones
 9 . . . . . . . . . X . . . . . X . . . 9
 8 . . . . . . . . . X O . . . . . . . . 8
 7 . . . . . . . O X O . O . . . . . . . 7
 6 . . . . . . . O X . . . . . . . . . . 6
 5 . . . . . . . X X . . . . . . X . . . 5
 4 . . O + O . . . O X X . . . . + . . . 4
 3 . . . . . . . O . O . X . . . X . . . 3
 2 . . . . . . . . . . O X . . . . . . . 2
 1 . . . . . . . . . . . . . . . . . . . 1
   A B C D E F G H J K L M N O P Q R S T

> As soon as we add a stone at M10, it is only weakly
> surrounded. As you can see, without the stone at M10,
> it is very difficult to find a single move that stops
> the dragon from escaping. This is a qualitative
> difference, and as such is appropriately measured by
> a criterion that is discrete, not continuous.

Sure. The escape measure is 3, and the weakness estimate 0.81. If we add
a white stone at M10, these numbers change to 7 and 0.59. Not completely
off, but it could also be better. The surrounded heuristic would
definitely help here. (...but...an improved escape value would help,
too...)

But now if you remove the stones around Q13/R13/R14 and instead place a
white stone at K16, then playing M10 won't make the white dragon
unsurrounded. But still it is the move here to play, trying to break out.

(Note that with your current implementation it doesn't matter much where
the white stone at K16 is placed, unless it is just on the boundary of
the convex hull.)

> >    A B C D E F G H J K L M N O P Q R S T
> > 19 . . . . . . . . . . . . . . . . . . . 19
> > 18 . . . . . . . . . . . . . . . . . . . 18
> > 17 . . . . . . . . . . . . . . . . . . . 17
> > 16 . . . + . . . . . + . . . . . + . . . 16
> > 15 . . . . . . . . . . . . . . . . . . . 15
> > 14 . . . . . . . . . . . . . . . . . . . 14
> > 13 . . . . . . . . . . . . . . . . . . . 13
> > 12 . . . X . . . . . . . . . . . X X . . 12
> > 11 . . . . . . . . O . . . . . . . . . . 11
> > 10 . . . X . . . . . O . . . . . X . . . 10
> >  9 . . . . . . . . . . . . . . . . . . . 9
> >  8 . . . . . . . . . O . . . . . a . . . 8
> >  7 . . . . . . . . . . . . . . . . . . . 7
> >  6 . . . X . . . . . . . . . . . X . . . 6
> >  5 . . . . . . . . . . . . . . . . . . . 5
> >  4 . . X X . . . . . O . . . . . X X . . 4
> >  3 . . . . . . . . . . . . . . . . . . . 3
> >  2 . . . . . . . . . O . . . . . . . . . 2
> >  1 . . . . . . . . . . . . . . . . . . . 1
> >    A B C D E F G H J K L M N O P Q R S T
> >
> > The stone at K2 means that K4 is not surrounded. If
> > you remove it, K4 is surrounded. Now if you add a
> > black stone at _a_, it is only weakly surrounded (it
> > should be _not_ surrounded according to your
> > explanations, but running your patch on it gives
> > weakly surrounded).

(I agree with what you say about K2.)

> The last point --- that adding to the enclosure makes
> the dragon less surrounded --- is more disturbing.

Maybe one could, instead of taking the neighbouring dragons, only take
the worms that are a neighbour of a worm that are part of our dragon.
This might at least reduce this problem.

Btw, another example:

> >    A B C D E F G H J K L M N O P Q R S T
> > 19 . . . . . . . . . . . . . . . . . . . 19
> > 18 . . . . . . . . . . . . . . . . . . . 18
> > 17 . . . . . . . . . . . . . . . . . . . 17
> > 16 . . . + . . . . . O . . . . . + . . . 16
> > 15 . . . . . . . . . . . . . . . . . . . 15
> > 14 . . . . . . . . . . . . . . . . . . . 14
> > 13 . . . . . . . . . . . . . . . . . . . 13
> > 12 . . . X . . . . . . . . . . . X X . . 12
> > 11 . . . . . . . . O . . . . . . . . . . 11
> > 10 . . . X . . . . . . . . . . . X . . . 10
> >  9 . . . . . . . . . . . . . . . . . . . 9
> >  8 . . . . . . . . . . . . . . . . . . . 8
> >  7 . . . . . . . . . . . . . . . . . . . 7
> >  6 . . . X . . . . . O . . . . . X . . . 6
> >  5 . . . . . . . . . . . . . . . . . . . 5
> >  4 . . X X . . . . . O . . . . . X X . . 4
> >  3 . . . . . . . . . . . . . . . . . . . 3
> >  2 . . . . . . . . . . . . . . . . . . . 2
> >  1 . . . . . . . . . . . . . . . . . . . 1
> >    A B C D E F G H J K L M N O P Q R S T

Surely the white stone at J11, while it is in the convex hull of the
black dragons at D12 and Q12, has excellent escape potential.

So maybe a refinement would be: If we have only one (and narrow) direction
to escape, then this doesn't help much, because it can be blocked with
one move.
This might be worth trying out, too.

Arend