Looking for triangulator/interpolator

[Grant Edwards]

I need to interpolate an irregularly spaced set of sampled
points: Given a set of x,y,z points, I need to interpolate z
values for a much finer x,y grid.

I tried using the scipy sandbox delaunay module, but the
interpolators don't work: the natural neighbor interpolator
produces a surface with "holes" in it: the interpolator returns
NaNs for no reason for certain regions within the convex hull
(the convex hull looks right, and the input Z values in that
region don't look any different that regions that work).

The linear interpolator just segfualts no matter what data I
try with it.

In the past i've used the Delny/libqhull module, but it doesn't
do interpolation, so I had to write the interpolator in Python
(which works but is very slow).

Is there any "off the shelf" module that does interpolation of
irregularly spaced data?

--
Grant Edwards grante Yow! LOOK!! Sullen
at American teens wearing
visi.com MADRAS shorts and "Flock of
Seagulls" HAIRCUTS!

[Scott David Daniels]

Grant Edwards wrote:

I need to interpolate an irregularly spaced set of sampled
points: Given a set of x,y,z points, I need to interpolate z
values for a much finer x,y grid.

I tried using the scipy sandbox delaunay module, but the
interpolators don't work: the natural neighbor interpolator
produces a surface with "holes" in it: the interpolator returns
NaNs for no reason for certain regions within the convex hull
(the convex hull looks right, and the input Z values in that
region don't look any different that regions that work).

Sounds like a great opportunity for you to contribute. The easiest
contribution would be to find as small a case as you can that
demonstrates the problem, fixing it and add a test case would be,
obviously, a greater contribution.

--Scott David Daniels
sc***********@a cm.org

[Grant Edwards]

On 2006-05-26, Scott David Daniels <sc***********@ acm.org> wrote:

I tried using the scipy sandbox delaunay module, but the
interpolators don't work: the natural neighbor interpolator
produces a surface with "holes" in it: the interpolator returns
NaNs for no reason for certain regions within the convex hull
(the convex hull looks right, and the input Z values in that
region don't look any different that regions that work).
Sounds like a great opportunity for you to contribute.

I'll try, but it looks like I'm going to be working all weekend
as it is.
The easiest contribution would be to find as small a case as
you can that demonstrates the problem, fixing it and add a
test case would be, obviously, a greater contribution.

--
Grant Edwards grante Yow! Hmmm... an arrogant
at bouquet with a subtle
visi.com suggestion of POLYVINYL
CHLORIDE...

[Grant Edwards]

On 2006-05-27, Grant Edwards <gr****@visi.co m> wrote:

On 2006-05-26, Scott David Daniels <sc***********@ acm.org> wrote:

I tried using the scipy sandbox delaunay module, but the
interpolators don't work: the natural neighbor interpolator
produces a surface with "holes" in it: the interpolator returns
NaNs for no reason for certain regions within the convex hull
(the convex hull looks right, and the input Z values in that
region don't look any different that regions that work).

Sounds like a great opportunity for you to contribute.

I'll try, but it looks like I'm going to be working all weekend
as it is.

OTOH, it looks like I'm screwed either way. My python
interpolator is so hopelessly slow it's useless in practice. It
can only process 4 points per second and I need to process
arrays of 10,000 to 50,000 elements. :(

The easiest contribution would be to find as small a case as
you can that demonstrates the problem, fixing it and add a
test case would be, obviously, a greater contribution.

--
Grant Edwards grante Yow! YOW!! Now I
at understand advanced
visi.com MICROBIOLOGY and th' new
TAX REFORM laws!!

[John Machin]

Grant> OTOH, it looks like I'm screwed either way. My python
interpolator is so hopelessly slow it's useless in practice. It
can only process 4 points per second and I need to process
arrays of 10,000 to 50,000 elements. :(

Pardon my utter ignorance of scipy, but are neither psyco nor pyrex any
use?
Cheers,
John

[Grant Edwards]

On 2006-05-27, Grant Edwards <gr****@visi.co m> wrote:

On 2006-05-27, Grant Edwards <gr****@visi.co m> wrote:

On 2006-05-26, Scott David Daniels <sc***********@ acm.org> wrote:

I tried using the scipy sandbox delaunay module, but the
interpolators don't work: the natural neighbor interpolator
produces a surface with "holes" in it: the interpolator returns
NaNs for no reason for certain regions within the convex hull
(the convex hull looks right, and the input Z values in that
region don't look any different that regions that work).

Sounds like a great opportunity for you to contribute.

I'll try, but it looks like I'm going to be working all weekend
as it is.

OTOH, it looks like I'm screwed either way. My python
interpolator is so hopelessly slow it's useless in practice. It
can only process 4 points per second and I need to process
arrays of 10,000 to 50,000 elements. :(

I found another module that claims to do what I want

http://www.cdc.noaa.gov/people/jeffr.../griddata.html

But, no matter what data I pass, I get either all zeros or all
NaNs back. :/

I'm 0 for 3 now.

--
Grant Edwards grante Yow! I'm ZIPPY!! Are we
at having FUN yet??
visi.com

[Travis E. Oliphant]

Grant Edwards wrote:

I need to interpolate an irregularly spaced set of sampled
points: Given a set of x,y,z points, I need to interpolate z
values for a much finer x,y grid.

How many x,y,z points do you have?

Did you try the fitpack function bisplrep in scipy? It can work well as
long as you don't have too many starting points.

Here is an example of how to use it
from numpy import rand, exp, ogrid
from scipy import interpolate
x = 2*rand(20)-1
y = 2*rand(20)-1
z = (x+y)*exp(-6.0*(x*x+y*y))

tck = interpolate.bis plrep(x,y,z,s=0 ,xb=-1,xe=1,yb=-1,ye=1)

xnew = r_[-1:1:70j]
ynew = r_[-1:1:70j]
znew = interpolate.bis plev(xnew,ynew, tck)
There is a buglet that is fixed in SVN scipy that means you need to
enter xb, xe, yb, and ye manually.

-Travis

[Robert Kern]

Grant Edwards wrote:

I found another module that claims to do what I want

http://www.cdc.noaa.gov/people/jeffr.../griddata.html

But, no matter what data I pass, I get either all zeros or all
NaNs back. :/

I'm 0 for 3 now.

I pointed you to

http://www.scipy.org/Cookbook/Matplo...ly_spaced_data

earlier. Does that not do what you want?

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco

[Grant Edwards]

On 2006-05-27, Grant Edwards <gr****@visi.co m> wrote:

I found another module that claims to do what I want

http://www.cdc.noaa.gov/people/jeffr.../griddata.html

But, no matter what data I pass, I get either all zeros or all
NaNs back. :/

Aaarrrggh. After some more sweating and swearing, it looks like
both the griddata module and the scipy.delaunay. nn_interpolator
do work as long as you pass the preferred brand of arrays to
them and specify the mesh/grid using the right scheme.

My problems seem to have been caused by the interaction of a
number of factors:

1) Gnuplot.py seems to like convert _some_ floating-point
arrays to integer values before plotting them -- this only
seems to happen when passing 2D arrays to splot(). That
was breaking some of my data.

2) Converting the arrays to nested lists prevents the rounding
to an integer problem but it apparently transposed the x/y
axis without my noticing. Then the "holes" in some of the
interpolated surfaces showed up. It turns out there were
NaNs in the input data that were causing the holes, but
because of the transposed x/y axis, I was looking in the
wrong place in the data.

3) Attempting to use the griddata module resulted in mixing
array objects from pylab, numpy, numeric, and scipy (some
of which may or may not be the same -- I can't keep track).
Mixing array types seems to have tripped up some
extensions. AFAICT, python code is happy with any of the
array types since they're pretty much the same if you go by
"duck" typing. But C/Fortran extensions only seem to work
with one sort or the other, and some python modules that
wrap those extensions will pass anything that quacks on
down to C/Fortran code, when then gets confused. Maybe.

4) Even when the arrays were OK, there are a couple
incompatible ways to specify a mesh/grid, and I picked the
wrong one in at least one case.

--
Grant Edwards grante Yow! MMM-MM!! So THIS is
at BIO-NEBULATION!
visi.com

[Grant Edwards]

On 2006-05-27, Robert Kern <ro*********@gm ail.com> wrote:

Grant Edwards wrote:

I found another module that claims to do what I want

http://www.cdc.noaa.gov/people/jeffr.../griddata.html

But, no matter what data I pass, I get either all zeros or all
NaNs back. :/

I'm 0 for 3 now.

I pointed you to

http://www.scipy.org/Cookbook/Matplo...ly_spaced_data

earlier. Does that not do what you want?

Yes, after installer newer versions of things and straightening out
some other issues.

--
Grant Edwards grante Yow! Are we on STRIKE yet?
at
visi.com