Re: Re : [Bug-gnubg] Odd/even effect: post I made on bgonline

[Joseph Heled]

The net was trained "to evaluate relative equities"  for moves by adding pairs of positions for the different moves, but also on some absolute positions to get the cube decisions right. 
So, it may be that the accuracy is less in positions far from a possible cube - but it will surprise me.

Unless I am mistaken GNUbg still is the strongest 0ply player around (at least it was so at the time), but I am not sure if that explains to odd/even discrepancy. 

-Joseph

On Tue, Aug 25, 2009 at 9:44 PM, Massimiliano Maini <address@hidden> wrote:

On BGonline Tom Keith advanced the hypothesis that this was due
to the training method: the NN has been trained to evaluate relative
equities (to rank moves) but not a lot to ebvaluate absolute ones,
which is bad for cube decisions, but probably not much can be done.
What is more surprising (at least to me) is that gnubg NN has been
"trained" to mimic its own 2ply and not its own 1ply.

Anybody with more NN knowledge wanting to comment on this ?
Joseph, can you confirm that this may be a cause ?

MaX..




----- Message d'origine ----
> De : Christian Anthon <address@hidden>
> À : Massimiliano Maini <address@hidden>
> Cc : "address@hidden" <address@hidden>
> Envoyé le : Samedi, 22 Août 2009, 0h37mn 25s
> Objet : Re: [Bug-gnubg] Odd/even effect: post I made on bgonline
>

> I believe this position class is about as bad as it gets.
>
> Christian.
>

> On Fri, Aug 21, 2009 at 4:20 PM, Massimiliano Mainiwrote:

> >
> > http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=46778
> >
> > Spooky !!
> >
> > MaX.
> >
> >
> >
> >
> >
> >
> >
> > _______________________________________________
> > Bug-gnubg mailing list
> > address@hidden
> > http://lists.gnu.org/mailman/listinfo/bug-gnubg
> >
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg







_______________________________________________
Bug-gnubg mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/bug-gnubg