[Computer-go] living by two false eyes

Mark Boon tesujisoftware at gmail.com
Mon Apr 4 20:02:27 PDT 2011


On Mon, Apr 4, 2011 at 4:12 PM, Petr Baudis <pasky at ucw.cz> wrote:
> On Mon, Apr 04, 2011 at 02:13:52PM -1000, Mark Boon wrote:
>> On Sun, Apr 3, 2011 at 9:08 AM, Petr Baudis <pasky at ucw.cz> wrote:
>> > Pachi has the same rules, and fails miserably in these cases, since the
>> > chance that the false eye owner will fill it before *all* outer
>> > liberties are filled is quite high. It did not occur to me yet how to
>> > fix this without incurring a large performance penalty.
>>
>> How about trying not to fill in a false eye unless one of the
>> surrounding chains has less than 3 liberties? I'm sure it's not
>> perfect, but may be a fairly good rule anyway.
>
> This goes against another deep worry, major misevaluations in semeais;
> they usually do involve plenty of false eyes and this will lead to even
> more distorted biases. Semeais occur far more frequently than living by
> false eyes (though Jean-Loup has found a couple of real examples when
> digging through automated playtesting game records).

That is certainly a possibility. So you'd have to do some careful
testing. But by far the majority of (real) false eyes have a chain
with two liberties. I'd actually be interested to see a semeai where
the correct play is filling a false eye when all surrounding chains
have 3 liberties or more. I think it would be extremely rare, if it
ever happens at all. But maybe other, more subtle biases could prove
to be destructive, like not filling the false eye but the extra (own)
liberty instead.

Two-headed dragons, as these are called, are so rare that I wouldn't
bother trying to deal with them properly. But if you think you need
to, then this is what I'd try first.

    Mark Boon



More information about the Computer-go mailing list