RE: [Bug-gnubg] Re: from money/dmp to match

[Albert Silver]

Perhaps I'm off track here, but the way it seems to me is that the
longer the match, the greater the edge for the stronger side. This is
obvious, but it leads to another point: why? One could try to calculate
the average length of 3-pt or 5-pt matches for example, in terms of
number of games, and then try to see how much average gain is obtained
from each game, separating the DMP games in the matches and inserting
the statistical values obtained from the separate DMP testing. In other
words, suppose the average 5-pt match length (and I'm making this up) is
3.8 games with 0.7 games being DMP games. The average percentage gain
from the normal games (3.1) with cube play would be totaled, and then
the percentages from the DMPs would be added. The reason for separating
them in matches is that the new nets do not only gain better checker
play, but also better evaluations leading to better cube decisions and
better checker play according to score. These don't appear in DMP games,
so they are clearly different. The DMP results from these matches would
be removed, as the idea is to see the average gain, and then the results
from the 1-pt matches would be put instead as they are statistically
more precise.

I'm not a mathematician (graduated in literature) so if it is all wrong
I won't be too surprised.

Another way would be to see how much average error is obtained not per
game, but per move since the average game could vary quite a bit.

                                                Albert


> -----Original Message-----
> From: address@hidden [mailto:bug-gnubg-
> address@hidden On Behalf Of neilkaz
> Sent: Saturday, February 22, 2003 12:00 AM
> To: address@hidden; gnubg
> Subject: [Bug-gnubg] Re: from money/dmp to match
> 
> Please note my new email address. !!
> 
> Off the top of my head..I don't know how to figure out this non
trivial
> problem.
> 
> I refer you to Jacob/Trice's book (Can a Fish taste twice as good?) or
to
> Zedah's article.
> 
> Sorry...neilkaz.. (I'll be gone on holiday until March 1)
> 
> thx...!!
> 
> 
> ----- Original Message -----
> From: "Joseph Heled" <address@hidden>
> To: "gnubg" <address@hidden>
> Cc: "Neil Kaz" <address@hidden>
> Sent: Friday, February 21, 2003 6:38 PM
> Subject: from money/dmp to match
> 
> 
> > Here is a question for you all, but especially for Neil.
> >
> > I have 2 players (2 nets in this case). Here are the money and DMP
> > (1-ptr) result when I play one against the other,
> >
> > money 0.57000 ppg (67.98 28.13 1.78  32.02 8.59 0.28)
> >
> > 1-ptr 71.07%
> >
> >
> > What do you expect the number of wins to be in a 3ptr, 5ptr etc? and
> Why?
> >
> > I will post the self play results later.
> >
> > And Here is another one
> >
> > Money 0.12005 ppg  (53.86 15.40 0.84  46.14 11.51 0.45)
> >
> > 1-ptr 55.69%
> >
> > -Joseph
> >
> >
> >
> 
> 
> 
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-gnubg