Delaunay Tesselation

[scidam]

In case of Scipy or matplotlib, they both invoke triangulation algorithm implemented in the qhull library. So, they are
almost the same, may be some preprocessing steps differ between scipy.spatial.Delaunay and matplotlib.tri.triangulation.

You can easily test efficiency of such triangulation,e.g.

import numpy as np
from scipy.spatial import Delaunay
data = np.random.rand(12000, 2)
def test():
    tri = Delaunay(data)
    tri.equations.shape

# in IPython

%timeit test

Output:
15.9 ns ± 0.243 ns per loop (mean ± std. dev. of 7 runs, 100000000 loops each)

Qhull library is implemented in C, and it is quite fast. I used scipy's triangulation (not too much yet),
and it didn't show any crashes. I think it is sufficiently robust.