Re: [Bug-gnubg] Even, odd and half plies (WAS: New Contact Net Error Rates)

[Joseph Heled]

Works for crashed as well, with w somewhere in the 0.4 - 0.5 range.
In race, 0ply beats any combination.

So, if we implement such a thing, we will need it to be per class. w may 
be per class, but that is not very important, as w=0.5 will do quite well.

-Joseph

Joseph Heled wrote:
> Nis wrote:
>
> It seems your idea has lot's of merit. Sorry for being skeptical at first.
>
> > > I will check that.
> >
> >
> >
> > Just out of curiosity: How?
>
>
> That's what benchmarks are for, and we have put a lot of work into 
> generating one.
> The benchmark is far from perfect, but still very useful.
>
>
> 107485 contact benchmark positions
>  0Ply error - 0.01079  (w = 0)
>  1Ply error - 0.00956  (w = 1)
>
>  w * ply1 + (1-w) * ply0
>
>   w = 0.5   error - 0.009176
>   w = 0.75  error - 0.009155
>   w = 0.625 error - 0.009066
>   w = 0.57  error - 0.009070
>   w = 0.67  error - 0.009071
>
> Conclusions:
>
>   - 1ply does play contact positions better after all.
>   - 1.5 ply is better yet. The best w is around 0.6
>
> I need to see if the same applies to crashed, and I wonder what the race 
> result would be. It takes time to do those 1ply evaluations ...
>
> -Joseph
>
>
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-gnubg