How to replace NaNs in array by other array

Question

I have 2 Numpy arrays with the same length

array([ 0.9737068 , NaN, NaN, ..., -0.64236529,
 -0.88137541, -0.78318609])
array([ 0.9 , 0.7643, 0.61, ..., -0.64236529,
 -0.88137541, -0.78318609])

In the first array I have NaN values, how can I replace these NaN values with values from the second array. In this example my third array would be:

array([ 0.9737068 , Nan => 0.7643, NaN => 0.61 , ..., -0.64236529,
 -0.88137541, -0.78318609])

• StonedTensor: Do you want a new array or to replace one of them?

Answer 1

Using Numpy, the following works by applying a Boolean mask to both arrays:

import numpy as np
x = np.array([0.9737068, np.nan, np.nan, -0.64236529, -0.88137541, -0.78318609])
y = np.array([0.9, 0.7643, 0.61, -0.64236529, -0.88137541, -0.78318609])
x[np.isnan(x)] = y[np.isnan(x)]

Results in

In[1]: x
Out[1]: 
array([ 0.9737068 , 0.7643 , 0.61 , -0.64236529, -0.88137541,
 -0.78318609])

N.B. Running with %timeit, this solution takes < 4μs in repeated runs, vs. the other two Numpy solutions (at time of writing this) which both take 20-25μs

Answer 2

Pythonic solution:

import numpy as np
def no_nans_arrays(array_w_nans,array_no_nans):
 return np.array([array_w_nans[i] if not np.isnan(array_w_nans[i]) else array_no_nans[i] for i in range(len(array_w_nans))])

Answer 3

If not numpy, you might find this function helpful:

def swap(A,B,i,j):
 TEMP_B = B[j]
 B[j] = A[i]
 A[i] = TEMP_B
 return A,B

Iterate over your first array, if element is NaN, then use swap.

Taken from this question.

Answer 4

res = np.array([ a1 if not np.isnan(a1) else b1 for a1,b1 in zip(a, b) ])

np.isnan returns False if the given element in the np.array is np.NaN

• StonedTensor: a more Pythonic solution would be if not np.isnan(a1) instead of if np.isnan(a1)==False