Re: [gnugo-devel] engine/influence.c

[Gunnar Farneback]

Teun wrote:
> Would it be possible to cast the influence spreading in the form
> of a 2-D partial differential equation, for instance heat diffusion?
> Spreading translates into diffusion. Permeability in position dependent
> conductivity.

It might be, but I'm unsure. One property I had in mind when designing
the influence function was to not allow the influence from a given
source to "bend around" obstructing stones, at least not too sharply.
I guess that heat diffusion doesn't really have this property.

> For partial differential equations there is a host of fast algorithms.

What does fast mean in this context? I have the impression that it
usually means that an algorithm is faster than some less sophisticated
solution to the problem (e.g. FFT compared to DFT) but doesn't really
say anything in absolute numbers. Some measure of the number of
computations per board vertex for such algortihms would be
interesting.

> An extra bonus is that the solution of a partial differential equation
> is orientation independent as opposed to the current influence
> spreading.

Is the ununrolled implementation really orientation dependent? As far
as I know the orientation dependence was caused by a buggy code
optimization.

In any case I'm all in favor of using a PDE-based influence function
if one can be found with appropriate properties.

/Gunnar