Expand numpy array to be able to broadcast with second array of variable depth

Question

I have a function which can take an np.ndarray of shape (3,) or (3, N), or (3, N, M), etc.. I want to add to the input array an array of shape (3,). At the moment, I have to manually check the shape of the incoming array and if neccessary, expand the array that is added to it, so that I don't get a broadcasting error. Is there a function in numpy that can expand my array to allow broadcasting for an input array of arbitrary depth?

def myfunction(input_array):
    array_to_add = np.array([1, 2, 3])
    if len(input_array.shape) == 1:
        return input_array + array_to_add
    elif len(input_array.shape) == 2:
        return input_array + array_to_add[:, None]
    elif len(input_array.shape) == 3:
        return input_array + array_to_add[:, None, None]
    ...

mozway · Accepted Answer · 2024-07-23 12:39:56Z

1

One option would be to transpose before and after the addition:

(input_array.T + array_to_add).T

You could also use expand_dims to add the extra dimensions:

(np.expand_dims(array_to_add, tuple(range(1, input_array.ndim)))
 + input_array
)

Alternatively with broadcast_to on the reversed shape of input_array + transpose:

(np.broadcast_to(array_to_add, input_array.shape[::-1]).T
 + input_array
)

Or with a custom reshape:

(array_to_add.reshape(array_to_add.shape+(1,)*(input_array.ndim-1))
 + input_array
)

edited Jul 23, 2024 at 12:39

answered Jul 23, 2024 at 12:25

mozway

267k13 gold badges56 silver badges106 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Alex V. Over a year ago

Awesome! Thanks for the multitude of solutions. Why does the transpose-version work? Does broadcasting begin at the last dimension?

mozway Over a year ago

Yes, broadcasting is aligned on the trailing dimensions, which would become the leading ones in the transpose.

Soudipta Dutta · Accepted Answer · 2024-08-05 13:20:20Z

To broadcast, Use the below formulae :

new_shape = (1,) *  (input_array.ndim -1) + array_to_add.shape

Code :

import numpy as np

array1 = np.array([[1, 2], [3, 4], [5, 6]])
array2 = np.array([10, 20])

print(array1.shape)#(3, 2)
print(array2.shape)#(2,)

print(array1.shape[-1])#2
print(array2.shape[0])#2


array_to_add = array2 
input_array = array1 

if  input_array.shape[-1] != array_to_add.shape[0] : 
    raise ValueError('Shapes are incompatable.')

new_shape = (1,) *  (input_array.ndim -1) + array_to_add.shape   

'''
new_shape :
(1, 2)
'''
array_to_add_expanded = array_to_add.reshape(new_shape)
'''
array_to_add_expanded :
[[10 20]]
'''
res = input_array + array_to_add_expanded
print(res)
'''
[[11 22]
 [13 24]
 [15 26]]
'''

Collectives™ on Stack Overflow

Expand numpy array to be able to broadcast with second array of variable depth

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related