How to remove NaN values from a given NumPy array?

In this article, we are going to learn how to remove Nan values from a given array. Nan values are those values that do not have a specific value associated with them or they are different from the type of values that are to be used in the declared array.

There are basically three approaches with slight differences in syntax. Either we could use a function specified in NumPy or we could use an operator, the basic working will be the same.

Using numpy.logical_not() and numpy.nan() functions

The numpy.isnan() will give true indexes for all the indexes where the value is nan and when combined with numpy.logical_not() function the boolean values will be reversed. So, in the end, we get indexes for all the elements which are not nan. From the indexes, we can filter out the values that are not nan and save them in another array.

import numpy
# create a 1D array
a = numpy.array([5, 2, 8, 9, 3, numpy.nan,
                 2, 6, 1, numpy.nan])
# remove nan values using numpy.isnan()
# and numpy.logical_not
b = a[numpy.logical_not(numpy.isnan(a))]
# print the results
print("original 1D array                    ->", a)
print("1D array after removing nan values   ->", b)
print()
# create a 2D array
c = numpy.array([[6, 2, numpy.nan], [2, 6, 1],
                 [numpy.nan, 1, numpy.nan]])
# remove nan values using numpy.isnan()
# and numpy.logical_not
d = c[numpy.logical_not(numpy.isnan(c))]
# print the results
print("Original 2D array   ->")
print(c)
print("2D array converted to 1D after removing nan values   ", d)

Output:

Note: No matter what the Dimension of the array is, it will be flattened into a 1D array

Using np.isnan() Remove NaN values from a given NumPy

Combining the ~ operator instead of numpy.logical_not() with numpy.isnan() function. This will work the same way as the above, it will convert any dimension array into a 1D array. 

import numpy
# create a 2D array
c = numpy.array([[12, 5, numpy.nan, 7],
                 [2, 61, 1, numpy.nan],
                 [numpy.nan, 1,
                  numpy.nan, 5]])
# remove nan values using numpy.isnan()
# and numpy.logical_not
d = c[~(numpy.isnan(c))]
# print the results
print("Original 2D array   ")
print(c)
print()
print("2D array converted to 1D after removing nan values   ")
print(d)

Output:

Using np.isfinite Remove NaN values from a given NumPy

The numpy.isfinite() function tests element-wise whether it is finite or not(not infinity or not Not a Number) and returns the result as a boolean array. Using this function we will get indexes for all the elements which are not nan. From the indexes, we can filter out the values that are not nan and save them in another array.

import numpy as np
# create a 2D array
c = np.array([[12, 5, np.nan, 7], 
                 [2, 61, 1, np.nan],
                 [np.nan, 1,
                  np.nan, 5]])
# remove nan values using numpy.isnan() 
# and numpy.logical_not
d = c[np.isfinite(c)]
# print the results
print("Original 2D array   ")
print(c, '\n')
print("2D array converted to 1D after removing nan values ")
print(d)

Output:

Original 2D array   
[[12.  5. nan  7.]
 [ 2. 61.  1. nan]
 [nan  1. nan  5.]] 
2D array converted to 1D after removing nan values 
[12.  5.  7.  2. 61.  1.  1.  5.]