[Scipy-tickets] [SciPy] #1666: LinearNDInterpolator fails when dimensions differ by many orders of magnitude

SciPy Trac scipy-tickets@scipy....
Sat Jun 2 10:57:57 CDT 2012


#1666: LinearNDInterpolator fails when dimensions differ by many orders of
magnitude
-----------------------------------------+----------------------------------
 Reporter:  bloop369                     |       Owner:  somebody   
     Type:  defect                       |      Status:  new        
 Priority:  normal                       |   Milestone:  Unscheduled
Component:  scipy.interpolate            |     Version:  0.10.0     
 Keywords:  LinearNDInterpolator, QHull  |  
-----------------------------------------+----------------------------------

Comment(by pv):

 The scattered data intepolation needs answer to the question "which data
 points are close to this one", and this requires a metric. The metric used
 in Qhull is ||x-y||_2. Obviously, the results are not affine-invariant
 (including scaling). Physically, using such a metric implies that all the
 dimensions must have the same units.

 Now, your suggestion of making the data axes dimensionless by scaling them
 with the range of the data is one possibility, but this may be the wrong
 choice in several cases (when the dimensions already have the same units).

 I believe the behavior as it is now is correct. AFAIK, it's also how
 things work in other numerical packages. This means the user must scale
 the data axes before interpolation in a sensible way.

 This can be spelled out in more detail in the documentation, and maybe
 options could be added to make scaling the data more easily controllable
 by the user.

-- 
Ticket URL: <http://projects.scipy.org/scipy/ticket/1666#comment:1>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list