[Yt-dev] clump finder stuff

[Stephen Skory]

Hi Matt,

> And it might be easier if you figured out why it was broken ratehr
> than waiting on me.  The issue is in the .pyx, not the .py file, and
> it is that the buffer of potential contour linkages is not long
> enough.  I thought it was, based on the loop structure that followed.
> If you expand it to the correct length it will work.


I've done some work on this, and you're right that the buffer in construct_boundary_relationships needs to be made bigger. In some cases, much bigger. The multiplier was set to 9:

s = (ny*nx + nx*nz + nx*nz - 4) * 9


In my dataset I've found that I need at least 52 for it to work in all cases. I did some statistics and the mean minimum multiplier required is 18 with a standard deviation of 9, so 52 is about 3 sigma out, so it's not that rare. I don't know how globally-applicable these statistics are; this is just one dataset.

What I'm wondering is what we should do about this? Make the multiplier large enough to capture (N)sigma likelihood? Or wrap the call to construct_boundary_relationships in countour_finder.py in a while loop with a try/except that increases the multiplier with each failure?


 
Stephen Skory
stephenskory at yahoo.com
http://stephenskory.com/
510.621.3687 (google voice)