Re: computer-go: Fast scoring program?

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

Hi Robert,

Ok, "spirit of  the game"  is very poor wording.   

In Chess tournaments between countries,  it has been a common practice
in the past to play a few moves (among your countrymen) and then agree
to a draw.  This is  a "game" itself, outside  of the real game and is
not "in the spirit of" chess.

At  the very beginning of  a GO  game, and  at any  point before it is
played out to the bitter end, every position  has a correct evaluation
but this is too complicated for humans to figure out most of the time.
At  some point before the  game  is "actually  over"  (see my  comment
below) it becomes obvious  to humans what the final result will be and
they agree on it.   This is less obvious to computers.

That's really what my "spirit of the game" comment was all about.

* Based on your comments, "actually over" means ignoring any rules
  that go beyond the mathmatical nature of the game, such as "player
  can by mutual agreement can do this or that."   

don




   Date: Thu, 16 Aug 2001 11:49:39 +0200
   From: Robert Jasiek <jasiek@xxxxxxxxxxxxxxxxx>

   "Grajdeanu, Adrian" wrote:
   > TT scoring does not capture the sprit of the game.

   Such a statement is pointless because you do not define
   "spirit of the game".

   > Counting dead stones as
   > points instead of prisonners and territory is bad.

   What is this supposed to say? That
   score = number of dead stones
   is worse than
   score = prisoners plus territory
   ?
   No doubt, the first does not make sense at all.
   Inhowfar are prisoners not points?
   To summarize, I do not understand your sentence at all.

   > Playing inside own territory is bad.

   In which context?

   Don Dailey <drd@xxxxxxxxxxxxxxxxx> wrote:
   > TT scoring does   indeed capture the  spirit  of  the game.

   Same remark as above for you.

   > Japanese
   > scoring is  just a shortcut way  to evalute the  end result before the
   > game actually ends.   

   Using the word "end" here is a little dangerous because it
   can have multiple meanings:
   - the game end (just before applying the score definition)
   - the state created by applying the score definition
   - the state after mechanically counting the score
   - the state after the result agreement

   > In my opinion, you should be training your NN  to understand the WHOLE
   > game, not just the part humans play  (which does not include that full
   > end game.)

   Similarly "play" as you use it can have even more meanings.

   > That is why I say TT captures  the true spirit of the game
   > and Japanese does not.

   Your point seems to be that "spirit" ends at the state created
   by applying the score definition. Why? Is it not necessary for
   a program to count the score, to reconfirm the result and the 
   winner, and to make a result agreement?

   Again let me ask: What is the spirit of the game? To put it
   differently: What is a game of go? When does ot start, when
   does it end, when do players/programs perform competition?
   A game starts from the empty board, then competition performs
   actions defined as legal by the rules, then the rules define
   termination of that performance so that their definition of
   the winner can be applied, then that definition is applied.
   We still need a context of tournament rules: Is it the task 
   of the rules of play to determine the winner or is the task of
   the players/programs? Depending on the answer, the final end
   of a game differs, the final end of skill differs, and 
   counting, reconfirmation, and result agreement may or may not
   be part of a game or of skill during it. So the "spirit of
   competition" is not given per se but only due to the priorly
   given rules of play together with the tournament rules. Even
   more confusingly, special rulesets like Japanese style rules
   may temporarily take any skill from the players to the rules
   themselves (during the scoring application) and thereafter
   give skill back to the players, who then may or may not be 
   able to interpret the score application, to count, to
   reconfirm, to agree on the result.

   > You are trying   to teach your  program  to  have subtle  end of  game
   > scoring  sense,  by using   the japanese  model  somehow, which   just
   > supplies the answer for it  and does not  let it make the mistakes  it
   > needs to learn with.

   For AI or NN programs different rulesets could be used for
   the score definition and independently of this different
   rulesets and their interpretation could be used as models
   for strategic evaluation during late stages of the game.
   E.g. my commentary on the Nihon Kiin 1989 rules is still
   useful for strategy if the program plays with TT rules for
   the score definition.

   --
   robert jasiek