Create boolean mask for an array of tuples

Question

I have a 2D numpy array with a shape (3, 3) and dtype=object whose elements are tuples of the form (str, str, float).

template = ('Apple', 'Orange', 5.0)
my_array = np.array([None] * 9).reshape((3,3))

for i in range(my_array.shape[0]):
    for j in range(my_array.shape[1]):
        my_array[i, j] = template

But when I try to get a boolean mask

print(my_array == template)

The answer is all False

[[False False False]
 [False False False]
 [False False False]]

However element-wise comparison still works

print(my_array[0,0] == template) # This prints True

Why does the boolean mask return all False and how do I make it work?

P.S. I have searched for relevant topics but couldn't make use of any...

Array of tuples in Python
Restructuring Array of Tuples
Apply function to an array of tuples
Filter numpy array of tuples

Answer 1

What is happening here is that in Python tuples are compared by position. So when you do

my_array == template

what you are actually doing (row-wise) is:

('Apple', 'Orange', 5.0) == 'Apple'
('Apple', 'Orange', 5.0) == 'Orange'
('Apple', 'Orange', 5.0) == 5.0

To verify that this is the case, try experimenting with the following example:

>>> other_array = np.array(['Apple', 'Orange', 5.0] * 3).reshape(3,3)
>>> other_array
array([['Apple', 'Orange', '5.0'],
       ['Apple', 'Orange', '5.0'],
       ['Apple', 'Orange', '5.0']], dtype='<U6')
>>> other_array == template
array([[ True,  True,  True],
       [ True,  True,  True],
       [ True,  True,  True]])

I don't know of any non-hackish way to work around this and get direct equality comparison working. If a hack suffices and your array is not too large you could try:

mask = np.array(list(map(lambda x: x == template,
                         my_array.flatten()))).reshape(my_array.shape)

or

mask = np.array([x == template for x in my_array.flatten()]).reshape(my_array.shape)

Is there a reason why you need a array of tuples? Can't you have another dimension in your array, or maybe use pandas for your categorical variables?