From: eaglesondouglas@gmail.com   
      
   On Monday, April 1, 2019 at 12:51:47 AM UTC-4, Jonathan Thornburg wrote:   
   > I have a data-analysis problem, and I wonder if someone can suggest   
   > software to solve it.   
   >   
   > Summary:   
   > I have a number of points (typically 300 to 1000) in $R^2$.  I'm looking   
   > for code (in any reasonable programming language) to find the smallest-area   
   > ellipsoid which encloses 95% of the points.  The code doesn't need to be   
   > particularly efficient.   
      
   [Moderator's note:  Rest of quoted post deleted. I'm not sure that the   
   following works.  It is clear what needs to be done; the question is how   
   to do it.  Future posts on this topic should ideally point to actual   
   code.  -P.H.]   
      
   Consider an ellipsoid as a set of line segments.  This is a graphical   
   approximation of a function.  This discrete representation can   
   be made as fine as desired.   
      
   Now for each line segment place an interior normal and exterior   
   normal pair of points.   
      
   For each point in your set calculate the distance to the   
   two normal's.  Do this for each line segment.  If the   
   distance to the interior normal is smaller than to the   
   exterior normal the point is inside the ellipsoid if   
   true for all line segments.   
      
   Now in the global sense graphically reduce or expand   
   the ellipsoid until the scoring of 95% for your set   
   of points is found.   
      
   This is a nonfunctional method, but can approximate   
   with as fine a resolution as wanted.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)