[Bug-gnubg] Re: Snowie error rates versus gnubg error rates

[Christopher D. Yep]

First, as a quick aside, let me note that this is my first post to the 
group.  On May 27, I downloaded GNU Backgammon (for Windows) for the first 
time (May 24 build).  I have been extremely impressed with all the features 
packed into GNU.  Also, as far as I can tell, the playing strength of GNU 
is comparable to Snowie 3.  From my small selection of positions tested, it 
is in fact stronger than Snowie 3.

Thank you to all who have worked on this software for so many years.  It is 
truly amazing that such a fine piece of software is free to download!

I subscribe to the daily digests, so I hope this response turns out fine 
(if not, let me know).  Hopefully you'll find this information useful.

I've used Snowie 3 for 3.5 years (since it's release).  I analyze all my 
online matches (500+ matches) as well as some of my live matches.  A number 
of years ago I did my own study of Snowie error rates.  I came up with the 
following Snowie error rate definitions:

Error rates for player 0:

Chequerplay error rate:  Em(0)/(m(0)+m(1))
Cube error rate:         Ec(0)/(m(0)+m(1))
Total error rate:        (Ec(0)+Em(0))/(m(0)+m(1))

I.e. the error rate for both chequerplay and cube has nothing to do with 
the number of cube decisions.  On the other hand it uses the total number 
of chequer plays for *both* players as the denominator.

Here m(0) = number of chequer play decisions for player 0,
m(1) = number of chequer play decisions for player 1
Em(0) = total normalised error for chequerplay decisions for player 0
Ec(0) = total normalised error for cube decisions for player 0

I tested the above definitions and they always matched exactly, up to 
round-off error, in Snowie 3 (I don't have Snowie 4 and never had Snowie 2).

Note that m(0) is approximately equal to m(1).  Also note that when player 
0 has cube access, m(1) is approximately equal to c(0) (i.e. my formulas 
essentially agree with your formulas when player 0 has cube access 100% of 
the time).

Can you check your formulas again?

Chris


PS: Thanks again for the great work you've done on GNU Backgammon!


At Sun, 06 Apr 2003 12:10:03 +0000, Joern Thyssen <address@hidden> wrote:
> Hi,
>
> I've studied snowie error rates versus gnubg error rates a bit:
>
> m=number of chequer play decisions (m'=number of unforced moves)
> c=number of cube decisions         (c'=number of non-trivial cube decisions)
>
> Em=total normalised error for chequerplay decisions
> Ec=total normalised error for cube decisions
>
>
> Snowie error rates:
>
> Chequerplay error rate:  Em/(m+c)
> Cube error rate:         Ec/(m+c)
> Total error rate:        (Ec+Em)/(m+c)
>
> gnubg error rates:
>
> Chequerplay error rate:  Em/m'
> Cube error rate:         Ec/c'
> Total error rate         (Ec+Em)/(m'+c')