[Bug-gnubg] Re: The importance of METs

[Joern Thyssen]

On Fri, Aug 29, 2003 at 05:17:52PM -0300, Albert Silver wrote
> I have a question that is really bothering me about this new MET and the
> importance of the difference between the METs in general. At first, I
> was quite enthusiastic about this new one, scoring better than others,
> because I read the difference as being 1.2% better instead of the actual
> 0.12% per match (DUH!). It's true this is an improvement but how
> significant is this, and why isn't the difference larger? I can *easily*
> show positions where just changing the MET will show an equity
> difference between doubling decisions to be an error or not. 
> That's in
> just one position mind you. So why isn't the overall effect on the match
> results larger? 

I guess the differences average out.

> Is the sample too small? I realize 500,000 matches is a
> lot, but perhaps the number of doubling decisions where a different MET
> would lead to a different decision is already small, and where this
> would actually make a difference in the score is smaller still, meaning
> that perhaps 500,000 is still much too small a sample. 

For a 0.12% difference you would need more than 1,000,000 games.

Jørn

pgpLqebswitzj.pgp
Description: PGP signature