Re: [pbmserv-dev] Those wacky provisionals

[Lyman Hurd <lhurd@yahoo.com>]

"so, if the opponent was established with a rating of
a little more than 2000"

Hmmm, the preceding email said that there wasn't such
a player.  There may be a slight problem with the
formula for small N.

Here's a code snippet from ratings.cpp simplified for
the case where there is 0 handicap and player2 is
established:

int rating1 = before.Rating(ix1);
value = effrat2 + 400*outcome;
newrating = (rating1*games1+value)/(games1+1) + 
            (1720-rating1)/5;

We assume effrat2 is the true rating of our
established opponent, and outcome = 1 (our newbie
won), the wildcard here is the assumed value of
rating1.  I plugged in an assumption of 1600 for
rating1 and cannot reach the observed levels.  Are we
sure that the initial rating (as passed in via before)
is correct?

I set up a spreadsheet to try out various values and I
don't think that the observed results are possible
unless rating1 is smaller!

But, provisional ratings don't mean much (at least for
the first few games).  This situation is exactly why a
provisional rating can go down even with wins!

I am more concerned about the use of ratings in
multiplayer games.  The formula used assumes a net
zero change in ratings when two players play. 
However, in a game with one winner and greater than
one loser this rule is violated!

Cheers,

Lyman

--- Richard Rognlie <rrognlie@gamerz.net> wrote:
> We all know the rating system (while you are
> provisional) can give 
> counterintuitive results.  but it works.
> 
> 
> so... we have a case where 1 player is provisional
> (no games completed)
> vs. 1 player whose status we don't know.
> 
> Let's assume the other player is also provisional.
> 
> The player gets an assumed rating of 1600.  the
> other player has a rating
> of N (say... 2000?).  Average them (1800).  And add
> 200 (2000).  *then* add the skew back towards 1720
> (1944).
> 
> not enough to get even close to the 2114 rating.  so
> assume that 
> that's not possible.
> 
> Now let's assume a game against an established
> player.
> 
> assumed rating 1600 vs. say 2000 (average 1800) +
> 400 (2200) 
> skew to 2104.  (so, if the opponent was established
> with a rating of
> a little more than 2000, it's possible that the
> rating would turn
> out to 21xx.
> 
> but it's a *provisional* rating.  It really doesn't
> matter.   
> 
> The established player loses nothing against a
> provisional player's
> first game. 
> 
> On Sun, Mar 30, 2003 at 03:38:27PM -0500, Marc
> Lanctot wrote:
> > The point was that it's a bug that should be
> fixed.
> > 
> > Marc
> > 
> > 
> > 
> > Martin Moller Pedersen wrote:
> > 
> > >[Charset iso-8859-1 unsupported, filtering to
> ASCII...]
> > >-> I just noticed that jon.dann has a provisional
> Quoridor rating of 2119 
> > >-> after playing one game.  Since the highest
> established rating is 1790, 
> > >and -> the second highest provisional is 1824, I
> am wondering how this 
> > >-> occurred.  I thought a provisional couldn't be
> more than 200 above the 
> > >-> player he beat.  This suggests that the rating
> system can be cleverly 
> > >abused.
> > >
> > >How can it cleverly be abused ?
> > >If a provisional player has a very high rating
> but has only played a 
> > >singled game
> > >and he then plays and lose a game against a
> established player, then the
> > >established player doesn't got more than a few
> points.
> > >
> > >/Martin
> > >
> > >
> > >To unsubscribe, send a message to
> esquire@gamerz.net with
> > >	unsubscribe pbmserv-dev@gamerz.net
> > >as the BODY of the message.  The SUBJECT is
> ignored.
> > >
> > > 
> > >
> > 
> > 
> > 
> > To unsubscribe, send a message to
> esquire@gamerz.net with
> > 	unsubscribe pbmserv-dev@gamerz.net
> > as the BODY of the message.  The SUBJECT is
> ignored.
> > 
> 
> -- 
>  /  \__  | Richard Rognlie / Sendmail Ninja /
> Gamerz.NET Lackey
>  \__/  \ | http://www.gamerz.net/rrognlie/   
> <rrognlie@gamerz.net>
>  /  \__/ | Today is the first day of the rest of
> your life.  Of course,
>  \__/    |    so was yesterday, and look how you
> screwed *that* up...
> 
> 
> To unsubscribe, send a message to esquire@gamerz.net
> with
> 	unsubscribe pbmserv-dev@gamerz.net
> as the BODY of the message.  The SUBJECT is ignored.
>