Re: [Bug-gnubg] Eval

[Joern Thyssen]

On Sat, Mar 08, 2003 at 03:03:53PM -0300, Albert Silver wrote
> 
> 
> > -----Original Message-----
> > From: address@hidden [mailto:bug-gnubg-
> > address@hidden On Behalf Of Joern Thyssen
> > Sent: Saturday, March 08, 2003 1:55 PM
> > To: Albert Silver
> > Cc: 'GNUBackgammon bug reporting'
> > Subject: Re: [Bug-gnubg] Eval
> > 
> > On Sat, Mar 08, 2003 at 01:18:45PM -0300, Albert Silver wrote
> > > There's something wrong with GNU's eval. Take the following position
> and
> > > look at the respective evaluations at 0, 1, 2, and 3 plies.
> > 
> > What's wrong? This is obviously a position that neural net doesn't
> > grasp, hence the huge difference between odd and even plies.
> > 
> > Are you playing with Jacoby rule? This explains why the 2-ply equity
> is
> > since -1, because gnubg doesn't see a single market gainer.
> 
> Yes, the Jacoby rule is on. Actually it shows the same evaluation if
> both cube and Jacoby are turned off, but shouldn't the eval not go
> beyond -1.000 if the Jacoby rule is on and the cube is in the center?

None of the evaluations has a cubeful "no double" equity lower than -1.

I don't see any problems or internal inconsistencies in the evaluations
as presented. Perhaps you can clarify?

However, the evaluations do look very different from what I get. Can you
reproduce them? 

Jørn