I have some 3D particle data
(x0,y0,z0) (x1,y1,z1) (x2,y2,z2) ...
I want to find the irregular bounding shape of the distribution. The image below shows an example distribution from three directions (x/y, y/z, x/z), as well as a spherical bounding box, which is a poor approximation to the true distribution.
For this particular use case, I wish to throw away particles near the edge of the distribution, since in my application these are subject to boundary effects. So, if I have a rough approximation of the bounding shape in 3D, I can scale this down and discard particles outside the re-scaled shape.
Solutions in interpreted languages (Python, R) preferable.