"jorma kala" <[EMAIL PROTECTED]> wrote on 28/02/2008 18:54:56:

> Sorry, one more question about the error rate:
> is the EMG equity difference used in the computation of the error 
> rate, is it  based on a cubeless evaluation?
> thanks.

Hi Jorma,
not sure to interpret correctly your questions, I don't even know if
you have 1 or 2 separate questions.

First of all, the errors used to compute the error rate are cubeful,
unless you have specified in the analysis settings that you do want
a cubeless analysis.

Concerning the overall error rate, for money sessions it's easy: it's
just the sum of the errors (no need to normalize) divided by the number
of unforced decisions.

For match-play, I suppose you have to proceed game by game: in each game
you add all the errrors in terms of MWC and then you convert into EMG 
(this
is equivalent to converting each MWC error into EMG and then sum, since 
the
MWC<->EMG mapping is linear).
Then, you just sum the EMG errors of all the games and divide by the total
number of unforced decisions (this is equivalent to computing the error 
rate
in each game and then taking the weighted average of the error rates of 
all
the games, weighted with respect to the number of unforced decisions).

Supposing EMG(x) is the function that converts from MWC to EMG (at a gven 
score):

Emwc(1)_i = i-th error in game 1, in terms of MWC
Eemg(1)_i = i-th error in game 1, in terms of EMG = EMG(Emwc(1)_i)

Emwc(1) = Total error in terms of MWC in game 1 = Sum(i,Emwc(1)_i)
Eemg(1) = Total error in terms of EMG in game 1 = Sum(i,Eemg(1)_i)
                                                = Sum(i,EMG(Emwc(1)_i))
                                                = EMG(Sum(i,Emwc(1)_i))
Nud(1)  = number of unforced decisions in game 1
ER(1)   = Error rate in game 1 = Eemg(1)/Nud(1)

ER      = Error rate of the match = Sum(j,ER(j)*Nud(j))/Sum(k,Nud(k))
                                  = Sum(j,Eemg(j))/Sum(k,Nud(k))

Notice that:
- the above holds for money sessions too, if MWC are replaced by equities 
and
if the function EMG(x) is replaced by the identity function.
- you cannot sum all the errors in terms of MWC across games and then 
convert
the total into EMG. The conversion into EMG is done at a given score (it 
depends
on the score). For money sessions, where the EMG function is replaced by 
the identity,
everything is simpler.


MaX.


P.S.
I post this to the mailing list: it may interest other users and I may 
have
written something wrong :))
_______________________________________________
Bug-gnubg mailing list
Bug-gnubg@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gnubg

Reply via email to