Array protocol and __array__ in NumPy - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output of this code using __array__?

Consider the following Python class that implements the __array__ method. What will be the output when converting an instance to a NumPy array?

NumPy

import numpy as np

class MyArray:
    def __array__(self):
        return np.array([10, 20, 30])

obj = MyArray()
result = np.array(obj)
print(result)

A [10 20 30]

B [[10 20 30]]

C

[[10]
 [20]
 [30]]

D TypeError: Cannot convert object to array

💡 Hint

The __array__ method returns a NumPy array representation of the object. NumPy uses this to create the array.

The __array__ method returns a 1D NumPy array [10, 20, 30]. When np.array(obj) is called, it uses this method to get the array data. So the output is a 1D array printed as [10 20 30].

❓ data_output

intermediate

What is the shape of the array returned by __array__?

Given this class with a __array__ method, what is the shape of the resulting NumPy array?

NumPy

import numpy as np

class Custom:
    def __array__(self):
        return np.array([[1, 2], [3, 4], [5, 6]])

obj = Custom()
arr = np.array(obj)
print(arr.shape)

A (1, 6)

B (2, 3)

C (6,)

D (3, 2)

💡 Hint

Look at the shape of the array returned inside __array__.

The __array__ method returns a 2D array with 3 rows and 2 columns, so the shape is (3, 2).

🔧 Debug

advanced

Why does this code raise an error when using __array__?

Examine the code below. Why does calling np.array(obj) raise an error?

NumPy

import numpy as np

class Broken:
    def __array__(self):
        return [1, 2, 'three']

obj = Broken()
np.array(obj)

A No error, returns array(['1', '2', 'three'], dtype='<U5')

B TypeError because the list contains mixed types that cannot be converted to a numeric array

C AttributeError because __array__ must return a NumPy array, not a list

D ValueError because the list length is inconsistent

💡 Hint

NumPy can create arrays from lists with mixed types by upcasting to strings.

The __array__ method returns a list with mixed types. NumPy converts this to an array of strings automatically, so no error occurs. The output is array(['1', '2', 'three'], dtype='.

🧠 Conceptual

advanced

What is the purpose of the __array__ method in NumPy?

Choose the best description of what the __array__ method does in a Python object used with NumPy.

A It automatically converts all object attributes to arrays

B It defines how the object converts itself to a NumPy array when passed to np.array()

C It is used to check if an object is already a NumPy array

D It prevents the object from being converted to a NumPy array

💡 Hint

Think about how NumPy uses this method internally.

The __array__ method tells NumPy how to get an array representation of the object. When np.array() is called on the object, NumPy calls this method to get the data.

🚀 Application

expert

Using __array__ to customize array conversion behavior

Given the class below, what will be the output of np.array(obj)?

NumPy

import numpy as np

class Wrapper:
    def __init__(self, data):
        self.data = data
    def __array__(self, dtype=None):
        arr = np.array(self.data)
        if dtype:
            return arr.astype(dtype)
        return arr * 2

obj = Wrapper([1, 2, 3])
result = np.array(obj, dtype=np.float64)
print(result)

A [2 4 6]

B [2. 4. 6.]

C [1. 2. 3.]

D TypeError because __array__ does not accept dtype argument

💡 Hint

The __array__ method can accept a dtype argument to customize conversion.

The __array__ method receives the dtype=np.float64 argument from np.array(). It converts self.data to an array [1 2 3], then since dtype is truthy, returns arr.astype(np.float64), resulting in [1. 2. 3.]. The multiplication by 2 only occurs if no dtype is specified.