Re: [gnugo-devel] persistent caching for break-in reading

[Gunnar Farneback]

Arend wrote:
> It should be pointed out that this does make a big difference even
> when examing only a single position. The reason is that the break-in
> analysis is repeated for every move that gets a territorial evaluation.

It would also be interesting to redo the false eye territory analysis
for each move. I was afraid that might be too expensive, but compared
to breakin analysis it should be significantly cheaper. Certainly less
important as well but in the small endgame it can make a few points of
difference.

> Actually, the connection reading shadow looks a little too big to me.

That's quite possible. In any case the persistent caching for
connections didn't turn out to be very effective.

/Gunnar