Finally had a chance to have alook at this data (thx
to work being a bit slow).

There is an inflection point (-19,0.5) that the data
revolves around, which applies to all data whether
there is 1 tile or 30 tiles left. A more realistic
estimate would be to move the inflection point
positively to(-5, 0.5). (playing 1 tile in the
endgame=4pts) Unless you have data that claims
otherwise.

Going by raw data, the win% ignores whether the
opponent has a Q or a blank.  I know this is
implemented to reduce runtime, but putting estimates
of damper tiles according to endgame letter
playability may be suitable.  Of course this means the
bogowin file would leap 1000 times in size: a dynamic
estimate would perhaps be more appropriate: even one
whereby quackle evolves as it plays endgames and
changes the damper values minutely according to end
result would provide a more precise estimate to win%.

Ranking plays by valuation in preference to win% would
be my preference until a more reliable win% system is
devised.  The difference between a 75% win and 80% win
decision is minute in comparison to a valuation. The
significance is greater when comparing 55% and 60%win
estimates.  Anything above 85% win is grossly appears
grossly unreliable, but I will look more into the
certainty limit factors when I have time.

I feel like i'm speaking an entirely different
language with the above, hope this makes sense.

Russell


--- jasonkatzbrown <[EMAIL PROTECTED]> wrote:

> --- In [email protected], Russell Honeybun
> <[EMAIL PROTECTED]> wrote:
> >
> > >Win % is an estimate of how often the play wins
> the 
> > >game, based on guess at the end of each
> simulation 
> > >iteration that is a function of difference in
> score 
> > >and how many tiles are left in the bag.
> > 
> > Could you please provide this equation in terms of
> > score_diff and tiles_left so I can calculate the
> > reliability of this estimation?
> 
> Look at 
> 
> quackle/data/strategy/twl06/bogowin
> 
> Format is
> 
> <spread> <tiles left> <win prob>
> 
> --Jason
> 
> 
> > 
> > 
> > --- jasonkatzbrown <[EMAIL PROTECTED]> wrote:
> > 
> > > --- In [email protected], "shaundonnelly"
> > > <shaundonnelly@> wrote:
> > > >
> > > > Hi,
> > > > 
> > > > I'm sure this has been asked before but i cant
> > > find an answer.
> > > > 
> > > > What do win percentage and valuation mean in
> terms
> > > of a given play and
> > > > how are they calculated.
> > > 
> > > Hi! Good questions.
> > > 
> > > Win % is an estimate of how often the play wins
> the
> > > game, based on
> > > guess at the end of each simulation iteration
> that
> > > is a function of
> > > difference in score and how many tiles are left
> in
> > > the bag.
> > > 
> > > Valuation is just score + leave before
> simulation,
> > > and (our scores -
> > > their scores + our rack value - their rack
> value)
> > > averaged over all of
> > > the simulation iterations.
> > > 
> > > > 
> > > > Also why was the term equity replaced with
> > > valuation, is there a
> > > > difference.
> > > 
> > > No, I just got tired of hearing the word
> "equity" =P
> > > 
> > > Cheers,
> > > Jason
> > > 
> > > 
> > > 
> > >  
> > > Yahoo! Groups Links
> > > 
> > > 
> > > 
> > > 
> > > 
> > 
> > 
> > 
> >        
> >
>
____________________________________________________________________________________
> > Moody friends. Drama queens. Your life? Nope! -
> their life, your
> story. Play Sims Stories at Yahoo! Games.
> > http://sims.yahoo.com/
> >
> 
> 
> 
> 
>  
> Yahoo! Groups Links
> 
> 
> 
> 
> 



      
____________________________________________________________________________________
Get easy, one-click access to your favorites. 
Make Yahoo! your homepage.
http://www.yahoo.com/r/hs 

Reply via email to