Re: [computer-go] Protocol B

[Ech Naton <ech_naton@xxxxxxxxx>]

Hello everybody


I do not know if this is of any help, but here is how
I implemented my protocol to do tournaments to test
different versions of my go:

Rules:
- chines
- superpositional
- invalid moves result in a pass

Game end:
- after two passes
- after 400 moves are reached
- if the two opponents agree on who owns which field
of the board

Counting:
- if the opponents agreed, count for each the fields,
that they agreed, each owns
- otherwise give 1 point for each stone still on the
board and for each field totaly surounded by only one
color


Commands to implement:
- Move  getMove(Game game)

Data structures:
Move: boardcoordinate or passconstant, array of
endboard or null)
Game: gamesize, startposition, moves up to now


Considerations:
- I do not support Japanese rules, to complicated and
inconsistend and unclear
- leave it to the engine to do a simple check if a new
game was sent or just a new move to the last game.
- if a engine passes, it must be aware of the fact,
that the opponent can finish the game and that all
stones, even dead ones will be counted.
- engines that can't remove dead stones, will lose :-)
- the engines do not agree on dead stones, but on who
gets which field of the board
- abandon the concept of dead stones, there are no
clear and simple rules to identify them
- i did not consider time issues for this protocol
- when I test against gnu go over gtp, i let it send
its endboard estimate with the move, when it passes.
If my program which is much weaker, then passes and
does not agree on the score, then it may win.

-> Maybe think about using a rulset that leaves the
things, that are not simple to the engines and let
them adapt if they like to participate in the
tournament. 

Regards Patrick 

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/