Re: [gnugo-devel] Semeai test suite

[Gunnar Farneback]

Ricard wrote:
> My name is Ricard Vilà. I've been working the last six months with 
> Tristan Cazenave in the developement of a semeai solver machine. I've 
> developed a complete semeai test suite following the conditions of the 
> go text protocol.

Great!

> I've defined my own gtp_semeai_solve function to use 
> with my program, and now i would like to compare the results achieved 
> with the answers that gnugo would give to the same problems.
> The input to my function are two coordinates of the two groups involved 
> in the semeai (the first coordinate sets the color to play first) and 
> the output is a semeai status (win, lost, seki, ko, unknown) plus the 
> intersection to play (it might be PASS).
> The problem is that, for what i've seen, there is no special function 
> for gnugo to deal with semeai. Do you know a way for me to perform the 
> tests i make to my program to gnugo? Maybe using owl_attack?

GNU Go 3.4 has a gtp function called analyze_semeai. Like your
function it takes two coordinates for the groups involved in the
semeai and the color of the first group plays first. Call the groups A
and B. The output differs though. GNU Go responds with two numbers and
an intersection to play (might be PASS here too). The first number
gives the outcome of defending A and the second number the outcome of
attacking B. The numbers are as follows:

0 fail
1 win, unconditional
2 ko, the opponent (compared to who moved first) must make the first
  external ko threat
3 ko, we must make the first external ko threat

Thus we have the following common correspondences
1 1 <intersection> - The player to move first wins the semeai
1 1 PASS           - The player to move first can tenuki and will
                     still win the semeai
1 0 <intersection> - Seki or both lives independently
1 0 PASS           - Seki or both lives independently regardless who
                     moves first
0 0 PASS           - The player to move first loses the semeai
2 2 <intersection> - Ko
3 3 <intersection> - Ko

It is worth noting that GNU Go currently always gives priority to seki
over ko when it has a choice between them.

> P.S. If you are interested in something i may provide you don't hesitate 
> to ask ;)

Uh, hey, we are *extremely* interested in a semeai test suite. Could
we have a copy, please?

/Gunnar