Why can't __array_function__ be called from an instance of an np.ndarray?

Question

I am trying to write a custom array container following numpy's guide and I can't understand why the following code always returns NotImplemented.

import numpy as np
a = np.array([1])
shape = (2, 2)
a.__array_function__(func=np.broadcast_to, types=(np.ndarray, type(shape)), args=(a, shape), kwargs={})
Out[5]: NotImplemented

Even though the normal method obviously works

np.broadcast_to(a, shape)
Out[6]: 
array([[1, 1],
       [1, 1]])

The function np.broacast_to is just an example, np.reshape has the same issue.

Does anyone know what my mistake is?

user2357112 · Accepted Answer · 2021-02-02 22:37:06Z

2

types is only supposed to include types that actually implement __array_function__. type(shape) doesn't implement __array_function__, so it shouldn't be in types.

answered Feb 2, 2021 at 22:37

user2357112

286k32 gold badges489 silver badges569 bronze badges

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

2 Comments

Roy Smart Over a year ago

That was definitely the problem, so how does __array_function__ know which type corresponds to which argument if the two iterables have different lengths?

Patol75 Over a year ago

Have a look here. The first few paragraphs before the example hold the answer.

Collectives™ on Stack Overflow

Why can't __array_function__ be called from an instance of an np.ndarray?

1 Answer 1

2 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Related