Replacing empty or missing values with zeros in a large array

Question

I have an large array of more than 40000 elements

a = ['15', '12', '', 18909, ...., '8989', '', '90789', '8']

I'm looking for a simply way to replace the empty '' values to '0' so that I can manipulate the data in the array using Numpy.

I would then convert the elements in my array into integers using

a = map(int, a)

so that I could find the mean of the array in numpy

a_mean = np.mean(a)

My issue is that I cannot convert to integers in an array with missing numbers to get a mean.

• Jon Clements: Can you do: new_a = [int(v or 0) for v in a] and then use new_a?
• Karn Kumar: I believe you can use numpy.nan_to_num ?

Answer 1

You could make a small function that converts a single value exactly how you want it, e.g.:

def to_int(x):
 try:
 return int(x)
 except ValueError:
 return 0

which can be used with map:

In [22]: a = ['15', '12', '', 18909, '8989', '90789', '8']
map(to_int, a)
Out[23]: [15, 12, 0, 18909, 8989, 90789, 8]

in a list comprehension:

In [25]: np.array([to_int(x) for x in a])
Out[25]: array([ 15, 12, 0, 18909, 8989, 90789, 8])

or in a generator expression to directly create a numpy array:

In [27]: np.fromiter((to_int(x) for x in a), dtype=int)
Out[27]: array([ 15, 12, 0, 18909, 8989, 90789, 8])

• user1821176: Thank you, this was the simple and clean solution to my problem.

Answer 2

If I understood you right so it should look like that:

for index in range(len(a)):
 if a[i] is '':
 a[i] = '0'

You can also use:

a = list(map(lambda x: '0' if x == '' else x, a))

Answer 3

From the previous learning with SO, i see you can impy the below solution to convert the NaN to zeros..

from numpy import *
a = array([[0, 1, 2], [3, 4, NaN]])
where_are_NaNs = isnan(a)
a[where_are_NaNs] = 0

secondly, nan_to_num() as i said earlier in my comment.

>>> import numpy as np
>>> a = array([[0, 1, 2], [3, 4, np.NaN]])
>>> a
array([[ 0., 1., 2.],
 [ 3., 4., nan]])
>>> a = np.nan_to_num(a)
>>> a
array([[ 0., 1., 2.],
 [ 3., 4., 0.]])

Answer 4

A more verbose answer is:

acc = 0
for v in a:
 acc+=int(v or 0)
a_mean = acc/len(a)