CODEWITHSUNDEEP
pythonnumpyscipyinterpolation
YohanRoth
YohanRoth

Reputation: 3263

Fill in missing values in 2d matrix using interpolation in numpy/scipy

I need to fill in missing values (given as 0) in 2d matrix. How would I accomplish it in numpy/scipy? I found scipy.interpolate.interp2d function but I cannot quite understand how to make it fill in zeros only without modifying non-zero entries.

Here is example of this function being used to smooth out the image https://scipython.com/book/chapter-8-scipy/examples/scipyinterpolateinterp2d/

but this is not what I am looking for. I just want to fill out zero values.

For example, the matrix is 

import numpy as np
mat = np.array([[1,2,0,0,4], [1,0,0,0,8], [0,4,2,2,0], [0,0,0,0,8], [1,0,0,0,1]])

mat
array([[1, 2, 0, 0, 4],
       [1, 0, 0, 0, 8],
       [0, 4, 2, 2, 0],
       [0, 0, 0, 0, 8],
       [1, 0, 0, 0, 1]])

In this matrix, all zeros must be replaced with interpolated values while original values should remain the same. What can I use for this task?

Upvotes: 2

Views: 2686

Answers (1)

stevemo
stevemo

Reputation: 1097

You have to make a decision how you want to fill in the zeros. For example, you could just use the average value in the array:

mat[mat == 0] = np.average(mat)
mat
# array([[1, 2, 1, 1, 4],
#        [1, 1, 1, 1, 8],
#        [1, 4, 2, 2, 1],
#        [1, 1, 1, 1, 8],
#        [1, 1, 1, 1, 1]])

or you could use the values from some function fitted to the nonzero values --- scipy.interpolate.interp2d uses a "spline" (think polynomial):

from scipy.interpolate import interp2d

ix = np.where(mat != 0)
f = interp2d(ix[0], ix[1], mat[ix].flatten(), kind='linear')
mat2 = mat.copy()
mat2[mat==0] = f(range(5), range(5)).T[mat==0]
mat2
# array([[ 1,  2,  3,  4,  4],
#        [ 1,  1,  1,  1,  8],
#        [ 4,  4,  2,  2, 11],
#        [ 4,  3,  2,  1,  8],
#        [ 1,  0,  0,  0,  1]])

although I think you will find this approach pretty finicky, especially for such a small dataset.

You could also have a look at other imputation approaches, like nearest neighbors, etc.

Upvotes: 2

Related Questions