On Fri, Jul 18, 2008 at 10:11 AM, Jeff Whitaker <[EMAIL PROTECTED]> wrote:
> I'd like to see griddata functionality and Ryan May's wind barb patch in
> 0.98.3.
OK things seem to be moving pretty fast right now on several fronts,
so we may want to wait until mid next week before pushing anything
out. Robert and I exchanged a few emails last night about his
delaunay code. We have two choices on the dependency -- require an
external scikits.delaunay that the user will install, or ship it
*internally* in matplotlib, eg as matplotlib.delaunay (he is not too
keen to see us repeat the mistakes we made with enthought.traits
installing it ourselves externally). I prefer the matplotlib.delaunay
solution since many win32, os x or naive users will not be comfortable
with the svn download and source install.
But Robert pointed out to me that one reason he has avoided pushing
this further, eg into scipy, is that there are known degenerate cases
arising from floating point precision issues that need to be worked
out
He wrote:
My apologies, it does not segfault; it just returns an impossible
triangulation.
def test_slightly_degenerate(self):
data = np.array([[-1, -1], [-1, 0], [-1, 1],
[ 0, -1], [ 0, 0], [ 0, 1],
[ 1, -1 - np.finfo(np.float_).eps], [ 1, 0], [ 1, 1],
])
tri = dlny.Triangulation(data[:,0], data[:,1])
I thought it possible that one of our developers may be interested in
working on this, and he gave the following advice:
Basically, Fortune's sweepline algorithm for Delaunay triangulation
simply needs to be replaced with an algorithm that can be formulated
using Jonathan Shewchuck's robust predicates:
http://www.cs.cmu.edu/~quake/robust.html
Divide and conquer or incremental insertion are reasonable candidates.
Personally, I am willing to include the code in its current imperfect
form in mpl with your griddata wrapper, provided that we make clear in
the docstring that there are known degenerate cases (eg include
Robert's example). Caveat emptor: this is an oft requested piece of
code and I think many users would like to have something that works in
most cases. But I think everyone would be well served by having a
bullet-proof algorithm and Robert won't have time to work on this in
the upcoming months, so perhaps one of us could spend some time
looking at following Robert's suggestion and incorporating Jonathan
Shewchuck's robust predicates into the implementation.
So my advice is: let's fold the delaunay code into matplotlib.delaunay
for 0.98.3 while providing copious warnings in your griddata
interface, and get to work improving the algorithm for future
releases, making sure we send Robert patches to the scikit.delaunay
module if we manage to do anything useful so his implementation will
remain the official branch as long as he wants it to be. If everyone
agrees with this plan, I'll assume you'll take the lead on importing
the code, and providing some examples/tests.
And we can hold for Ryan's wind barbs too -- it looks like Eric is on the case.
JDH
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Matplotlib-devel mailing list
Matplotlib-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
