RE: [computer-go] KGS Tournaments: Cheating

["Persson, Magnus" <persson@xxxxxxxxxxxxxxxxxx>]

As an author of a program that do not play reproducible, I must state
that it is not necessarily easy to make such a program play
reproducibly.

In order to have random move generator (which is a necessity in my
program) providing the same result every time, I would need to set a
fixed limit of calls to the random generator at each move. But then I
lose control over time it takes to compute the move and would end up
having to play moves faster than the time limits allow. Now it also
happens to be the case that the playing strength of my program depends
strongly on the thinking time.

Thus, forcing a program to play in a reproducible way will be unfair to
it, because it would not be able to use the time controls in a flexible
and efficient way. It is impossible to know in advance how long time it
takes to compute the best move.

Magnus (author of viking5)

--
Magnus Persson
Center for Adaptive Behavior and Cognition
Tel: +49-(30)-82406-350
Cell phone: +49 163 6639868
> -----Original Message-----
> From: Arend Bayer [mailto:arend.bayer@xxxxxxxxxxxxxxxxx]
> 
> On Mon, 7 Mar 2005, Mark Boon wrote:
> 
> > On Mon, 7 Mar 2005 13:25:04 +0000, Nick Wedd <nick@xxxxxxxxxxxxxxxxx>
wrote:
> 
> > > Another proposal is that programs must play reproducibly.  Then,
if
> > > there is an accusation of cheating, the accused program can be
asked
> to
> > > reproduce the sequence of moves.  However, some programs don't
play
> > > reproducibly, and can't be forced to.
> > >
> 
> > Of course there could still be programs that can't reproduce play
for
> > other reasons than a random generator. One source could be bugs.
> > Forcing moves to be reproducible may actually be doing a favour to
> > those programs :) Another source for irreproducible moves is
> > time-control. If a program adjusts the amount it reads based on real
> > time used, you may occasionally get different moves when trying to
> > reproduce the game. This is be a little harder to crack. You may end
> 
> It depends on how the program implements the time control, of course.
If
> the program just adjusts its depth setings etc. according to the time
> spent on previous moves, then this shouldn't be a problem: it should
do
> so according to the "time_left" reports it gets via GTP from kgsgtp.
> Just logging the GTP stream should make everything reproducible.
computer-go/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/