Re: [Bug-gnubg] Can we display Effective Pip Count

[Joern Thyssen]

On Mon, May 12, 2003 at 10:27:43PM +0100, Ian Shaw wrote
> 
> ----- Original Message -----
> From: "Joern Thyssen" <address@hidden>
> 
> I've added effective pip count to
> 
> Analyse->One chequer race
> Analyse->One sided rollout
> Analyse->Evaluation (if one- or two-sided bearoff)
> 
> Jørn
> 
> Cool! What is "One sided rollout" anyway? I've never figured out what it
> does. Whatever it is, it's very quick. 

It's similar to a one-sided bearoff database, except that the bearoff
distributions are not calculated exact, but rather through a one sided
rollout.  That is, instead of taking all possible 21 rolls at each turn
until all your chequers are borne off, the OSR only does, say, 1296
trials. gnubg calculates are bearoff distribution for each side and
combines to yield wins, gammons, and backgammons completely analogous to
the use of one-sided bearoff databases.

The code is based on Joseph's code in fibs2html.

You can read a bit on

http://pages.quicksilver.net.nz/pepe/ngb/index-top.html#2ply%20Vs.%20OSR%20racing

I think Joseph at one point used one-sided rollouts instead of the race
net, since OSR is much faster than 2-ply (but slower than 0-ply), but as
it turned out 2-ply plays better than OSR.

> The Std Deviations never seem to have any values in them.

No, the current code doesn't calculate them... It's on my TODO list :-)

Jørn