RE: [Bug-gnubg] (no subject)

[Ian Shaw]

> -----Original Message-----
> From: address@hidden
> [mailto:address@hidden Behalf 
> Of Øystein
> O Johansen
> Sent: 16 July 2003 14:33
> To: address@hidden
> Cc: olivier croisille; address@hidden
> Subject: Re: [Bug-gnubg] (no subject)
> 
> 
> 
> I really agree with Olivier here. 4 digits is a bit too much. 
> In a normal
> 2-ply evaluation the 4th digit isn't significant.
> 
> However, in a rollout the 4th digit may be significan and 
> relevant. Can the
> MoveList and CubeAnalysis first check if it's a rollout or a 
> evaluation and
> then report 3 or 4 digits accoring to this?
> 
> -Øystein
> 

I started off this discussion because gnubg-nogui was reporting win 
percentages, std errs, etc to only 3 significant figures (sf), compared to 4 sf 
in the normal program. This particularly affected the calculation of the 95%CI.

I can't remember the precision of the overall equity, so it obviously didn't 
bother me.

I agree that in the normal run of things, 4 sf is overkill - over the board I'd 
be very happy if I could be accurate to 2! 

However, when doing theoretical work such as developing a MET, the experts are 
requesting accuracy to 3 sf. This means that we need to display 4 sf so that we 
don't get rounding errors. One of the main criticisms of the Woolsey-Heinrich 
MET is that it is to only 2 sf, which leads to large errors when calculating a 
takepoint. See http://mail.gnu.org/archive/html/bug-gnubg/2002-12/msg00306.html 
for a comment by Jørn on this issue.

I do have sympathy with Olivier & Øystein's view, but would be sad to lose the 
current precision. How would people feel if the percentages and std errors were 
reported to 4 sf, but the overall equity were to 3 sf?




--Ian