Changing element in 2D numpy array to nan

sample = np.array([[-1, 1, -1, 1], [-1, 2, -1, 2], [-1, 3, 3, 3] ,[-1, 4, 4, 4], [-1, 5, 5, 5], [6, 6, 6, -1], [7, 7, 7, -1], [8, 8, -1, -1], [9, 9, -1, -1]])
float_sample = sample.astype(np.float64)

for row in float_sample:
    for ele in row:
        if (ele == -1.):
            float_sample[row][ele] = np.nan

I am trying to iterate through a 2D numpy array and whenever I see -1, convert it to NaN, np.nan. But whenever I try to do it doing the iteration I have above I get the following error message:

"Arrays used as indices must be of integer (or boolean) type"

How to I fix it so that I can iterate through a 2D numpy array of type float and whenever it finds a -1, convert it to NaN? I am doing this so that I can take the median of each column but you can’t do that using masked arrays so I am stuck trying to do it with a normal numpy array.

>Solution :

Use numpy’s boolean indexing:

import numpy as np

sample = np.array(
    [[-1, 1, -1, 1], [-1, 2, -1, 2], [-1, 3, 3, 3], [-1, 4, 4, 4], [-1, 5, 5, 5], [6, 6, 6, -1], [7, 7, 7, -1],
     [8, 8, -1, -1], [9, 9, -1, -1]])

float_sample = sample.astype(np.float64)
float_sample[sample == -1] = np.nan
print(float_sample)

Output

[[nan  1. nan  1.]
 [nan  2. nan  2.]
 [nan  3.  3.  3.]
 [nan  4.  4.  4.]
 [nan  5.  5.  5.]
 [ 6.  6.  6. nan]
 [ 7.  7.  7. nan]
 [ 8.  8. nan nan]
 [ 9.  9. nan nan]]

As a side note the problem is that you are using the row to index in:

float_sample[row][ele] = np.nan

and row is an array of floats.