Re: [computer-go] Tesuji solver

[Arend Bayer <arend.bayer@xxxxxx>]

Hi,

> You can read much more about the algorithm etc. on the web pages. What might be of interest to other programmers is that Madlab comes with 997 tesuji problems extracted from real high-level games. These are a mix of normal and double-threat tesuji's, and the problems are available for download (sgf). The quality is very uneven, but some of them are quite amusing. And what is more: the solution given in the files is guaranteed to work. [That is, in the absence of some undiscovered bug in my program :-)]. So the problems might be useful for regressions, testing, etc.
> 

How long does your solver take to analyze these 997 problems? For the
record, GNU Go at level 15 takes 48 seconds, fails at 36--84 of the
434 normal attack tests [1], and 424 of the 563 double-threat
problems (which is not so surprising, since the "attack-either" analysis
in GNU Go is pretty crude).

To run the tests, I generated a GTP-file (".tst"), which I can share if
someone else is interested.

Arend


[1]: There are 48 tests where GNU Go finds a different answer than your
solver, and I haven't checked whether they are valid alternative solutions.
Would you be interested in trying to collect all valid solutions?


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/