Re: [computer-go] random evaluation results

[Gunnar Farnebäck <gunnar@xxxxxxxxxxxxxx>]

Don wrote:
> Would anyone be willing to verify my results using the same
> parameters?  Gunnar is doing something similar, but he is using a
> whole different methodology which would take me some time to
> duplicate.

Yes, my methodology differs in the following respects:
* 9x9 board.
* Baseline is the Brown algorithm.
* Instead of estimating win probability I try to determine fair
  handicap by adjusting it by one after each game, depending on who
  wins, and averaging the used handicaps. (Handicaps +1 and -1 are
  even games as white and black respectively, 0 is not used.)

Basically Brown is random play that does not fill own eyes
and does not pass until eye-filling moves are the only ones left. For
details on the algorithm, download
http://www.lysator.liu.se/~gunnar/gtp/brown-1.0.tar.gz
and read the documentation and code comments. One of the main features
with having this algorithm as baseline is that all games end in a
fully settled state. Stones with at least two liberties are
invincible (Benson-alive) and stones in atari are unconditionally
dead (seki can never occur). As a consequence all scores are odd and
komi 0 is equivalent with 0.5 and -0.5.

With Don's notation I use these parameters:

      boardsize:   9
	   komi:   0
	     ko:   simple
      game over:   after 3 passes (but not inside search)
	   eval:   gg_urand() (32 bit uniform distribution from GNU Go.)
	scoring:   Chinese - dead stones resolved

My results so far:

Depth  handicap  #games
5        2.70      50
4        1.29     728
2       -4.21    5004
3       -5.94    5101
0      -13.90    5006
1      -14.18    5033

Sample games can be downloaded from the "red - depth x" entries on
http://www.lysator.liu.se/~gunnar/gtp/random_go_challenge.html

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