[r-t] Predicting search size (was FKM)
mark at snowtiger.net
Wed Jan 9 20:21:49 UTC 2008
> Now I'm confused again. Surely you don't update this computation every
> time your search has to backtrack?
As I said, you can check the source code yourself! No, from memory I have a
counter, maybe every million backtracks or so the % complete is updated.
It's only worth calculating % complete if you need to display it to the
user. The point is, you can compute on any backtrack, it doesn't have to be
a complete composition.
> When I tried what I believe is the identical algorithm it really wasn't
> useful for me at all. While monotonic, it was horribly non-linear, and
> conveyed almost no useful information. I wonder if my aggressive pruning
> means there are more or deeper holes to fall into or whatever?
Quite possibly. As I mentioned, I have a search at the moment with a Big
Deep Hole in, where the % complete has obviously gone completely
pear-shaped; but generally it's pretty useful. Probably my faster node-rates
smooth out some variation, too.
More information about the ringing-theory