[SciPy-user] 2D Interpolation

Ryan May rmay31@gmail....
Fri Jun 27 15:46:30 CDT 2008


Pauli Virtanen wrote:
> Hi,
> 
> Fri, 27 Jun 2008 14:13:22 -0400, Ryan May wrote:
>> Can anyone help me use scipy.interpolate correctly.  Here's my problem:
>> I'm trying to make a 2D lookup table to save some calculations.  The two
>> parameters over which the lookup table is generated are independent and
>> I have complete control over how I divide up the domain.  Using this
>> lookup table, I'd like to then calculate values over an unstructured set
>> of parameter values (ie. a list of pairs of parameter values).  Is there
>> a function in scipy.interpolate that can help here? What I'd really like
>> to be able to do is generate an interpolator object from my 2D array,
>> and then pass a pair of 1D arrays to the object and have it return 1D
>> array of values.
> 
> I don't think there are currently any functions that do that, but 
> certainly we'd like to have them.
> 
> I created a Scipy enhancement ticket for this feature:
> 
> 	http://scipy.org/scipy/scipy/ticket/693
> 
> Attached to it is a quick patch that implements the necessary loop on the 
> Fortran side. I suspect the patch needs further work, as there possibly 
> are faster ways to vectorise this piece of computation than simply 
> calling fpbisp at each point separately. (If someone more familiar with 
> the spline code wants to bless it, I can commit it, though...)
> 
> Currently, you can use
> 
> 	spl = scipy.interpolate.RectBivariateSpline(xi,yi,zi)
> 	z = scipy.array([spl(xp, yp)[0,0] for xp, yp in zip(x, y)])
> 
> The Python overhead isn't as bad as it looks like; moving the loop to
> the Fortran side gains only a factor of 5 improvement in speed.

Thanks for that.  I'm surprised that the Python looping overhead isn't 
worse. I just automatically assumed looping of Python == death of my 
code.  I'll see what I get then just manually looping.

Thanks

-- 
Ryan May
Graduate Research Assistant
School of Meteorology
University of Oklahoma


More information about the SciPy-user mailing list