Matplotlib triangulation stops working on switching coordinates

Question

I get the following weird error when I switch x and y coordinates on a set of points

  File "min_working_example.py", line 8, in <module>
    triang = tria.Triangulation(coords[:,1],coords[:,0])
  File "/usr/lib/pymodules/python2.7/matplotlib/tri/triangulation.py", line 72, in __init__
    dt = delaunay.Triangulation(self.x, self.y)
  File "/usr/lib/pymodules/python2.7/matplotlib/delaunay/triangulate.py", line 123, in __init__
    self.hull = self._compute_convex_hull()
  File "/usr/lib/pymodules/python2.7/matplotlib/delaunay/triangulate.py", line 158, in _compute_convex_hull
    hull.append(edges.pop(hull[-1]))
KeyError: 0

I have attached the set of points here and a minimum working example which can generate this error. This is with matplotlib 1.3 on ubuntu 13.10 with the version from the repositories. Strangely enough, only this data set seems to be problematic.

Thanks, Subramanya

Code:

from pylab import *
import matplotlib.tri as tria

coords = loadtxt("data.txt")

fig = figure()
ax = fig.add_subplot(111)
triang = tria.Triangulation(coords[:,0],coords[:,1])
ax.triplot(triang)
savefig('a.png')
triang2 = tria.Triangulation(coords[:,1],coords[:,0])

Data

-6.500000000000004663e-01 -8.660254037844383745e-01

-6.500000000000003553e-01 -6.062177826491067512e-01
-6.500000000000003553e-01 -3.464101615137751833e-01
-6.500000000000002442e-01 -8.660254037844355990e-02
-6.500000000000001332e-01 1.732050807568880635e-01
-6.500000000000000222e-01 4.330127018922196869e-01
-6.499999999999999112e-01 6.928203230275513658e-01
-6.499999999999998002e-01 9.526279441628826561e-01
-4.250000000000004885e-01 -9.959292143521041307e-01
-4.250000000000003220e-01 -7.361215932167725073e-01
-4.250000000000002665e-01 -4.763139720814408840e-01
-4.250000000000001554e-01 -2.165063509461093161e-01
-4.250000000000000444e-01 4.330127018922230731e-02
-4.249999999999999334e-01 3.031088913245539307e-01
-4.249999999999998224e-01 5.629165124598856096e-01
-4.249999999999997669e-01 8.227241335952174550e-01
-2.000000000000003442e-01 -8.660254037844382635e-01
-2.000000000000002331e-01 -6.062177826491067512e-01
-2.000000000000001499e-01 -3.464101615137751278e-01
-2.000000000000000389e-01 -8.660254037844355990e-02
-1.999999999999999556e-01 1.732050807568881190e-01
-1.999999999999998446e-01 4.330127018922196869e-01
-1.999999999999997613e-01 6.928203230275513658e-01
 2.499999999999980016e-02 -7.361215932167725073e-01
 2.499999999999985567e-02 -4.763139720814409950e-01 
 2.499999999999999445e-02 -2.165063509461093716e-01
 2.500000000000007772e-02 4.330127018922225179e-02
 2.500000000000017486e-02 3.031088913245538752e-01
  2.500000000000029976e-02 5.629165124598854986e-01
 2.499999999999998335e-01 -6.062177826491067512e-01
  2.499999999999999722e-01 -3.464101615137752388e-01
  2.500000000000001110e-01 -8.660254037844355990e-02
  2.500000000000002220e-01 1.732050807568880080e-01
  2.500000000000002776e-01 4.330127018922196314e-01
  4.749999999999999778e-01 -4.763139720814410505e-01
  4.750000000000000888e-01 -2.165063509461093993e-01
  4.750000000000001998e-01 4.330127018922216853e-02
  4.750000000000003109e-01 3.031088913245537642e-01
  7.000000000000000666e-01 -3.464101615137752388e-01
  7.000000000000001776e-01 -8.660254037844367092e-02
  7.000000000000002887e-01 1.732050807568879525e-01
  9.250000000000002665e-01 -2.165063509461093716e-01
  9.250000000000004885e-01 4.330127018922221710e-02
  1.150000000000000355e+00 -8.660254037844382358e-02 

Answer 1

I can confirm this bug with matplotlib v1.3 (matplotlib.delaunay was known to be subject to those problems) but not with current master now using the more robust library qhull.

https://github.com/matplotlib/matplotlib/pull/2504

So this should soon be solved with a v1.4 release I guess.

Below the graph obtained with current master.