From: Tim Casey <tcasey <at> MV <dot> US <dot> ADOBE <dot> COM>
Date: 13 dec 1995
Subject: Re: IGS/ISS thread

> From: Albyn Jones <jones <at> REED <dot> EDU>

> there should be no significant differences in the rating algorithm.
> one question is: what is the nominal difference between ranks?  in
> go i believe that a one stone difference in rank is presumed to
> correspond to a winning probability of 2/3 for the higher ranked player.
> is the traditional ranking system in shogi based on the same assumption?

Well, this is already a significant difference in the algorithm.  There
is no idea about 'stones'.  (At least the last time I played Shogi :) :).

Yes, the basic presumption about the rating system is the basis for difference
in rank.  One stone is the same as one rank, or there abouts.

So for Shogi games, a handicap game should not be counted.  Because I do
not know of any piece to rank difference off hand.  I imagine there is one,
just I am not sure how simple it would be.  Also, in Go, the person who
goes first gets a 0.5 advantage, where 1.0 == one stone.  None of this
makes sense for Shogi.  Although, the person who does has the first move
probably does have an advantage.

All of this comes down to I do not know how the rating system will perform
for Shogi games.  If I limit the ratings to even games, and a few other
things which come along, then we will see.  This took about 1 1/2 years
on IGS, just to set expectation up front :).

On the other hand, there is nothing inherent in the ratings about Go.
All it takes is the win/loss records for the day.  This sill happens.
The basic algorithm is a very sound one for computing ratings.  It is
used by the AGA and IGS.  The basic function of the ratings, collecting
results, adjusting people, etc; are all working.  The only question in
my mind is the question of picking up Go and putting in Shogi represents
a totally new set of unexpected results.  It could very well be things
will work.

So when someone comes along as a 4d, in shogi, and plays enough games;
their rating will be computed.  Now, if their ratings gets computed to
be a 1d, this is bad.  If it gets computed to a 5d/3d, maybe things
are working but need adjustment.  I guess this is the kind of process
I expect.  (Not the 4d->1d, but the 4d->5d or 4d->3d :).

tim