Fill in missing values with nearest neighbour in Python numpy masked arrays?

Question

I am working with a 2D Numpy masked_array in Python. I need to change the data values in the masked area such that they equal the nearest unmasked value.

NB. If there ar

Answer 1

You could use np.roll to make shifted copies of a, then use boolean logic on the masks to identify the spots to be filled in:

import numpy as np
import numpy.ma as ma

a = np.arange(100).reshape(10,10)
fill_value=-99
a[2:4,3:8] = fill_value
a[8,8] = fill_value
a = ma.masked_array(a,a==fill_value)
print(a)

# [[0 1 2 3 4 5 6 7 8 9]
#  [10 11 12 13 14 15 16 17 18 19]
#  [20 21 22 -- -- -- -- -- 28 29]
#  [30 31 32 -- -- -- -- -- 38 39]
#  [40 41 42 43 44 45 46 47 48 49]
#  [50 51 52 53 54 55 56 57 58 59]
#  [60 61 62 63 64 65 66 67 68 69]
#  [70 71 72 73 74 75 76 77 78 79]
#  [80 81 82 83 84 85 86 87 -- 89]
#  [90 91 92 93 94 95 96 97 98 99]]

for shift in (-1,1):
    for axis in (0,1):        
        a_shifted=np.roll(a,shift=shift,axis=axis)
        idx=~a_shifted.mask * a.mask
        a[idx]=a_shifted[idx]

print(a)

# [[0 1 2 3 4 5 6 7 8 9]
#  [10 11 12 13 14 15 16 17 18 19]
#  [20 21 22 13 14 15 16 28 28 29]
#  [30 31 32 43 44 45 46 47 38 39]
#  [40 41 42 43 44 45 46 47 48 49]
#  [50 51 52 53 54 55 56 57 58 59]
#  [60 61 62 63 64 65 66 67 68 69]
#  [70 71 72 73 74 75 76 77 78 79]
#  [80 81 82 83 84 85 86 87 98 89]
#  [90 91 92 93 94 95 96 97 98 99]]

---

If you'd like to use a larger set of nearest neighbors, you could perhaps do something like this:

neighbors=((0,1),(0,-1),(1,0),(-1,0),(1,1),(-1,1),(1,-1),(-1,-1),
           (0,2),(0,-2),(2,0),(-2,0))

Note that the order of the elements in neighbors is important. You probably want to fill in missing values with the nearest neighbor, not just any neighbor. There's probably a smarter way to generate the neighbors sequence, but I'm not seeing it at the moment.

a_copy=a.copy()
for hor_shift,vert_shift in neighbors:
    if not np.any(a.mask): break
    a_shifted=np.roll(a_copy,shift=hor_shift,axis=1)
    a_shifted=np.roll(a_shifted,shift=vert_shift,axis=0)
    idx=~a_shifted.mask*a.mask
    a[idx]=a_shifted[idx]

Note that np.roll happily rolls the lower edge to the top, so a missing value at the top may be filled in by a value from the very bottom. If this is a problem, I'd have to think more about how to fix it. The obvious but not very clever solution would be to use if statements and feed the edges a different sequence of admissible neighbors...