[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [Bug-gnubg] Importance of pipcount
From: |
David Montgomery |
Subject: |
RE: [Bug-gnubg] Importance of pipcount |
Date: |
Mon, 24 May 2004 10:38:43 -0700 |
Another issue with this experiment is that the pipcount
is probably a linear combo of the "raw" checker encoding.
So the net has access to the pipcount even if it isn't
explicitly represented.
> On Sun, May 23, 2004 at 06:45:10PM +0200, address@hidden wrote
> > I actually toyed with the idea of testing it out by hacking GNUBG to
> > get
> > a wrong pipcount input to its neural net and letting it play against
> > a
> > non-disabled version of itself and measure the effect.
>
> A few comments on this:
>
> * as far as I remember the pip count isn't used at all as input for
> the neural nets
>
> * if the pip count had indeed been an input we could have found the
> expected change in the cubeless money equity by using 'eval' which
> prints the derivative of the cubeless money equity. This is useful for
> finding out the effect of changing a given input, i.e.,
> delta equity ~= dE/dI * (delta input).
>
> Jørn