Re: [computer-go] Chains and liberties - performance

[Gunnar Farnebäck <gunnar@xxxxxxxxxxxxxx>]

Anders wrote:
> As far as I understand, GNU Go only keeps track of the liberties up to a
> certain number, not the theoretical maximum possible. Why? You'll have to
> test for the upper limit in a number of places, and recompute the liberties
> if you ever get blocks with more liberties. Is the reduced memory usage seen
> as more important than a small speedup?

Yes, it's a memory consideration. When the code was implemented it
made a difference of 650 kB, which I didn't think was justified to
spend. Now it's four and a half years later, GNU Go's memory footprint
is larger anyway, and the difference has shrunk to 210 kB. It may be
time to revise that decision.

The main reason to change would not be some tenths of a percent in
speed (my guess), however, but a reduction of the code complexity.

/Gunnar
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/