Re: [gnugo-devel] Surrounding moyo sizes

[Arend Bayer]

On Sun, 8 Sep 2002, Gunnar Farneback wrote:

> After this patch I think --experimental-connections is starting to
> become a solid option. The regression delta is 62 passes against 33
> failures. When playing --experimental-connections against
> non-experimental connections, the experimental option won 59/100 as
> white and 54/100 as black.

That sounds obvously great.

> The remaining issue is speed. Although I haven't measured it recently
> there's reason to suspect that it's still clearly slower with
> experimental connections. On the other hand there's no persistent
> cache for connections implemented yet.

Would it make sense to move out handling of persistent caches
into a separate file persisent_cache.c? Although the semantics of the
corresponding functions in reading.c and owl.c differ in details, I
think there is enough duplicated code to make this worth doing.

(Some parts, in particular computation of the the relevant active area, will
of course remain separate.)
E.g. I considered adding a hash table to the persistent reading cache
so that is size could be increased to 1000. That could be useful for
the connections, too.

Arend