If I have two arrays as shown below:

a = numpy.array([0, 0, 1, 0, 1, 1, 1, 0, 1])
b = numpy.array([1, 1, 1, 0, 0, 1, 1, 0, 0])

Is there an easy way using numpy to count the number of occurrences where elements at the same index in each of the two arrays have a value equal to one. In the above two arrays, the elements in position(zero-indexed) 2, 5 and 6 are equal to 1 in both the arrays. Thus I want to get a count of 3 here.

Thank you for any help that you may be able to provide.

  • $\begingroup$ This question is not about data-science, it is purely about code and belongs in stackoverflow. $\endgroup$ – Mark.F Jan 14 '19 at 12:06

There are two ways I'll show you (there are probably a lot more using NumPy):

First method: chaining operations

You can use "masking" followed by the comparison and finally a sum operation:

We want all values in a from the indices where b is equal to 1:

part1 = a[b == 1]

Now we want all places where part1 is equal to 1

part2 = part1[part1 == 1]

now we are left with all the places where a and b are equal to 1, so we can simply sum them up:

result = part2.sum()

Method 2: built in numpy.where

This is much shorted and probably faster to compute. NumPy has a nice function that returns the indices where your criteria are met in some arrays:

condition_1 = (a == 1)
condition_2 = (b == 1)

Now we can combine the operation by saying "and" - the binary operator version: &.

part1 = numpy.where(condition_1 & condition_2)

To get your desired output, we can take the length of the resulting set of indices:

result = len(part1)

Read the documentation about numpy.where to see the other things it can do for you!

sum(a * b) 

Should do the job:)

As pointed out by @n1k31t4 it only works if you have two arrays that contain only 0 and 1. Otherwise you would have to write something like:

 sum((a == 1) * (b ==1)) 

What I find interesting here is that the sum functions treats boolean vectors as vectors with 0 (for False) and 1 (for True) on which you can perform arithmetic operation (+ for or, * for and etc..)

  • $\begingroup$ This is an elegant solution, but only happens to work because OP used 0 and 1 in his example ;) $\endgroup$ – n1k31t4 Jan 14 '19 at 13:46

I like @RobinNicole's answer - in terms of Mathematics you are looking for a product of two boolean vectors.

Here are a few Numpy ways to do that:

In [37]: np.dot(a, b)
Out[37]: 3

In [38]: a @ b
Out[38]: 3

Here is another more generic solution which will work also for not-boolean vectors:

In [48]: ((a == 1) & (b == 1)).sum()
Out[48]: 3
  • $\begingroup$ This is an elegant solution, but only happens to work because OP used 0 and 1 in his example ;) $\endgroup$ – n1k31t4 Jan 14 '19 at 13:46
  • 1
    $\begingroup$ @n1k31t4, absolutely! That's why I marked word "boolean" as italian ;) I've also added a more generic Numpy solution $\endgroup$ – MaxU Jan 14 '19 at 13:48

Not the answer you're looking for? Browse other questions tagged or ask your own question.