def test_mesh_gaussian_value(self):
"""Check reference values and empty cases for mesh_gaussian"""
# Numerical values were not checked against an external reference
# so they are only useful for detecting if the results have _changed_.
self.assertEqual(
Broadening.mesh_gaussian(sigma=5, points=[], center=1).shape,
(0, ))
zero_result = Broadening.mesh_gaussian(sigma=np.array([[5]]),
points=np.array([
0,
]),
center=np.array([[3]]))
self.assertEqual(zero_result.shape, (1, 1))
self.assertFalse(zero_result.any())
assert_array_almost_equal(
Broadening.mesh_gaussian(sigma=2,
points=np.array([0, 1]),
center=0), np.array([0.199471, 0.176033]))
assert_array_almost_equal(
Broadening.mesh_gaussian(sigma=np.array([[2], [2]]),
points=np.array([0, 1, 2]),
center=np.array([[0], [1]])),
np.array([[0.199471, 0.176033, 0.120985],
[0.176033, 0.199471, 0.176033]]))
def test_mesh_gaussian_sum(self):
"""Check sum of mesh_gaussian is correctly adapted to bin width"""
# Note that larger bin widths will not sum to 1 with this theoretical normalisation factor; this is a
# consequence of directly evaluating the Gaussian function. For coarse bins, consider using the "normal" kernel
# which does not have this error.
for bin_width in 0.1, 0.35:
points = np.arange(-20, 20, bin_width)
curve = Broadening.mesh_gaussian(sigma=0.4, points=points)
self.assertAlmostEqual(sum(curve), 1)