Re: computer-go: Live or Die

[Compgo123@xxxxxxx]

I worked on life & death probelms for a while. I used exhaustive search for 
the maximum score (not only alive) with a limit on the maximum depth and 
tried to detect terminal positions as early as possible. Although a lot of 
progress was made, it's still far away from the results of Tom Wolf's 
program. Later I realized that Tom Wolf's results can only be obtained by 
selective search. Apparently he found a set of selection rules that will give 
the correct results probably > 99 percent of the time.

L&D is an area that most likely a mathematical theory will be found. One 
question is very interesting. What determines the complexicity of a L&D 
problem? I think it's a good idea if we can put forward several benchmark 
problems so that people can compare results with each other. Following is 
some problems I tried.

Problem 1:

Black: a4,b4,c2,c3,c4
White: a5,b5,c5,d1,d2,d5,e1,e2,e3,e4,e5

about 3 second with 1600 evaluations. Depth limit = 8.

Problem 2:

Black: b4,b5,c2,c3,c4,d2
White: a7,b6,b7,c5,c6,d3,d4,d5,e2,e3,f1,f2

81 second with 32000 evaluations. depth limit = 9


Problem 3:

Black: b4,c4,d2,d3,d4
White: a6,b5,b6,c5,d5,e2,e3,e4,e5,f1,f2

Note that this is not the Carpenter square.

80 minutes with 1.5 million evaluations. Depth limit = 12


Black moves first in all above problems. My computer is K-6 233 MHz.


Dan Liu