Re: [Bug-gnubg] Current categorization of doubling mistakes (new thread)

[Jim Segrave]

On Tue 07 Oct 2003 (07:00 +0000), Joern Thyssen wrote:
> On Tue, Oct 07, 2003 at 04:17:01AM +0200, olivier croisille wrote
> > 
> > Why on earth did we change categories in the first place anyway? I do favor 
> > change, and obviously improvements take time to implement, but here I 
> > cannot see *any* value-added over the former categories that were working 
> > sooooo fiiiiiiiine, without any mistake in 2 years+ in categorizing 
> > doubling mistakes. 
> 
> The original code I implemented simply had:
> 
> if ( cubeful equity > 0.95 )
>    around too good
> else
>    around double point
> 
> This code fails miserably for match play.
> 
> 
> The problem is how to define the "around" radius. 
> 
> J?rn

This leads to a new question on what must be becoming Joern's least
favourite subject...

I understand the idea of early/late doubling (I don't apply it well,
but I do understand it). Now, when gnubg analyses a game where I was 
creeping up on an optimal double (or never saw it was possible to
double or whatever), how does gnubg decide if not doubling was an
error or correct. Let's say 

DP = 75% CP = 81% TG = 84%

If I'm at 77% and don't double, is that an error of say .030? I assume
gnubg isn't examining market losers to decide if you should postpone
doubling? 


-- 
Jim Segrave           address@hidden