[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Computer Go Tournament Program



   From: Robert Jasiek <jasiek@xxxxxxxxxxxxxxxxx>

   Don Dailey wrote:
   > This is unduly  complicated and  solves no  problems.

   I know. Therefore I try to convince you how difficult
   application of such a protocol can become.

   > The whole point of this is utter simplicity.

   For this reason I prefer not to use any agreements before
   scoring because they create many more possibilties of
   disagreements. Agreements oppose simplicity.

   --
   robert jasiek


Ok, so you were playing "devil's advocate",  I was beginning to wonder
if something was wrong with you!

I have    always agreed  on this  point,   but there  are   some minor
advantages to such  a protocol if it  can be implemented  in a trivial
and purely optional way.

But maybe it cannot.  As I think more about it, I see a problem that I
hoped to avoid.  If  a program has pure  TT rules implemented, it must
be modified   at least slightly  to accomodate  this special agreement
protocol.  My hope was that a pure TT  program could remain completely
unmodified, essentially ignoring the protocol  if it chooses to.   But
this is not this case.

First  of  all, a  pure TT program   thinks there  only  has  to  be 2
consecutive  passes.  When the opponent offers   an agreement, it will
either  come  to the conclusion that   the opponent is doing  a stupid
thing and jump at the pass opportunity, or refuse  it outright but for
the wrong reason.  I'm not sure if either of these scenarios hurts the
programs final score, but at the very least they  are just plain ugly.
In either case, the program needs to know that it might take more than
2 passes to end the game.

I tried to determine  if the arbiter  software itself could compensate
in some way, (such as providing an implied  pass or pretending certain
passes  don't exist) but  even if this  is possible, each player has a
separate view of exactly what happened.  More ugliness.

I now believe that the  only way this  agreement protocol can work, is
if  all Tromp/Taylor programs   (assuming  others exist) are  slightly
modifed, which  means   the  original clean  and  simple  Tromp/Taylor
algorithm  could not be used  in its purest form.   The mod's would be
very  slight,  but already I don't  like  it,  and I won't  attempt to
implement the agreement protocol unless  someone  can talk me into  it
again!


Don