> Quote - And Cage, please don't mess with the threshold! > > You square your values, you compare them to the > squared threshold. It's really that simple. > > The distribution of values is only relevant when > you calculate weights. Of course squared distances > will lead to different weights than distances. I can > never remember how these things *should* work, but > if you want, I can have a look at your weight-calculating > code and try to remember my maths.

Okay.  This squared business is counter-intuitive.  :lol:
I tested not leaving it alone, and the results weren't helpful.  But they were
also confused by the problem I report, below....

I'm working on a big cleanup effort, but I've noticed a really bizarre bug.
This is present in the currently posted form of the comparison script,
and I have no idea what is causing it.  I'm guessing some kind of
float accuracy issues, but I can't track that down.

The problem: I'm consistently generating completely different sets of
correlations when using the pyd than when I'm not.  I don't see where this could
be happening, unless either the .pyd or Numeric is introducing float
accuracy errors somewhere.  The trouble, however, is that the .pyd results are
the good ones.  The non-.pyd results aren't very good at all, in my test case.

The problem is not accuracy issues due to squared distances.  I've removed the squared
distances from my WIP version of the script, thanks to odf's speed improvements (:woot:).
The problem I'm reporting, however, still remains.  It shouldn't be in the sort-handling
code, because then it would affect a .pyd run, as well.  In fact, I've altered most of the
close_verts_falloff() function, and the problem still carries over exactly as it is in the
current posted version of the script.

===========================sigline======================================================

Cage can be an opinionated jerk who posts without thinking.  He apologizes for this.  He's honestly not trying to be a turkeyhead.

Cage had some freebies, compatible with Poser 11 and below.  His Python scripts were saved at archive.org, along with the rest of the Morphography site, where they were hosted.