On Tue, 2008-10-21 at 10:53 +0200, Denis fidaali wrote:
> Then it'll be easy to be sure that this is conforming indeed.
> (although i'm not very happy about the super ko check
> maybe i'll skip it, and just pretend it is there, as that probably
> won't be noticed by the black-box test)

The black-box test will definitely test this situation.  In fact I would
probably make a specific test by asking the engine how many legal moves
are possible from some position or ask the program to list them.   Is
there a GTP command for that?    Another way to test it is to ask the
program to return a move where the "clearly best move" is PSK illegal
just to see if the program tries to play it.   

The bottom line is that I will try to build a test that you won't easily
fool.

A stage 2 test will be to play games against a reference bot and you
will probably fail this test if you don't know PSK.   You will have to
score within 2 standard deviations in a 10,000 game match of the
expected 50% result.  

But why sweat it?   Positional superko is trivial to implement and
requires no noticeable resources.  You don't have to check for superko
except one time, when you make a move.   So you could simply make a copy
of each board in the master games and compare them against the board
position you are about to create.   You could have a switch to turn this
off or use a different ko rule such as situational super ko (I think
that is what you said you prefer.)    But PSK is pretty much the most
widely used, for better or for worse.  

I would be happy if you made a java bot, please do that!

- Don

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to