Re: [gnugo-devel] Tripod group

[Gunnar Farneback]

Dan wrote:
> Here's the position.
> 
> +-----------
> |..*........
> |...X.......
> |.XXOOOOO...
> |..OO.......
> |..O........
> |..O........
> |..O........
> 
> Surprising, but it seems to me that an attack by O at C19
> can produce at least a ko.
> 
> I have in front of me a volume of Tsume Go problems by Cho
> Chikun (Chinese edition, I think) who considers a very
> similar position on p.98 and does not consider the move at
> C19.

If the contents are similar to his All About Life and Death (Ishi
Press), the discussion of the C19 move is postponed to the section
with this position

> +-----------
> |..*........
> |...XXO.....
> |.XXOOOOO...
> |..OO.......
> |..O........
> |..O........
> |..O........

where C19 does give ko. (The response to C19 in the original tripod
group is also discussed in that section and B19 is recommended.)


> However I cannot find a refutation better than
> ko. For a while I thought that pushing along the side at
> E18 worked, but GNU came up with B19 leading to this position:
> 
> +-----------
> |.OO........
> |...XX......
> |.XXOOOOO...
> |..OO.......
> |..O........
> |..O........
> |..O........
> 
> I can't find anything better than ko for black.

Blocking at F18 instead of playing B19 definitely gives ko and looks
more natural.

> The solution B18 given by Allan and Arend is convincing. There's
> no double ko.

B18 looks fine to me too.

> GNU Go prefers B19 which leads to variations looking like
> the bottom right corner in tripod8.sgf. I assume the
> double ko you are referring to is in this position.
> 
> +--------
> |O.O..... 
> |.OXX.... 
> |OXXOOOOO
> |.*OO.... 
> |..O..... 
> |..O..... 
> |..O..... 

Yes, or a few moves later. The final double ko life position can be
found in ld_owl:199.

> There's an actual error in GNU Go's reading, below
> the owl branch depth. Here's an edited variation
> tree. Note that GNU thinks the move is to the left of
> *, which actually produces ko. This is at some depth
> in the move tree. In the actual position GNU makes
> the right move at *, but we can't say it read it out
> correctly from the outset.

The tripod group is difficult to read. Last time I tuned for them GNU
Go went wrong in several variations in most of the tripod test cases.

> I think B18 is a much cleaner answer and we should tune
> GNU Go to play that way.

That's not the move recommended by Cho Chikun, but it's possible he
recommends B19 because he thinks the double ko life variation is more
instructive or for some other pedagogical reason.

Just make sure not to break ld_owl:182.

/Gunnar