Description
Description
#4206, once merged, exposes interesting behavior for measure.profile_line. Consider
image = np.arange(12).reshape(4, 3)
# reduce_func is a new parameter proposed in #4206
prof = profile_line(image, (0, 1), (3, 1), linewidth=3, reduce_func=None)
assert_array_equal(image, prof)As I understand it, the returned profile should be identical with the original image but we get this (value at (0, 2) is different):
AssertionError:
Arrays are not equal
Mismatch: 8.33%
Max absolute difference: 2
Max relative difference: 0.
x: array([[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8],
[ 9, 10, 11]])
y: array([[ 0, 1, 0],
[ 3, 4, 5],
[ 6, 7, 8],
[ 9, 10, 11]])
This might originate in scipy.ndimage.map_coordinates. Looks like a potential bug to me but might be an edge case for the underlying interpolation that can be ignored/is not reasonable to fix.
This behavior is already relevant even before merging #4206 because the faulty value will change the mean value of the profile line.
#4206 (comment) also highlights additional strange behavior.
Version information
Current master, exposed once #4206 is merged.