Inconsistent behaviour with numpy broadcasting while creating array of array

Question

If I'm trying to execute :

a = np.ones((1, 2))
b = np.ones((1, 3))

np.array([a, b], dtype=np.ndarray)

I get the following error :

ValueError: could not broadcast input array from shape (2,) into shape (1,)

But I'm would like to get :

array([array([[1., 1.]]), array([[1., 1., 1.]])], dtype=object)

But if I'm executing this :

a = np.ones((1, 2))
b = np.ones((2, 4))

np.array([a, b], dtype=np.ndarray)

I get the expected results :

array([array([[1., 1.]]), array([[1., 1., 1.],
                                 [1., 1., 1.]])], dtype=object)

Running on :

numpy==1.21.2 python==3.7.11

Make a 2 element object dtype array, and assign those 2 arrays to its slots. np.array can't be used directly for this. — hpaulj
– hpaulj, Commented Oct 15, 2021 at 8:01

Louis Hulot · Accepted Answer · 2021-10-15 08:12:20Z

1

Based on @hpaulh comments, this works:

import numpy as np


a = np.ones((1, 2))
b = np.ones((1, 3))

test = np.empty((2,), dtype=np.ndarray)

test[0] = a
test[1] = b

returns

array([array([[1., 1.]]), array([[1., 1., 1.]])], dtype=object)

answered Oct 15, 2021 at 8:12

Louis Hulot

3953 silver badges14 bronze badges

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

1 Answer 1