Numpy array not working when return array with single item

Question

Because plain dict is not adequate for inheritance, I design the following MyDict with UserDict in python standard libraries:

import numpy as np
from collections import UserDict

class MyUserDict(UserDict):
    pass

m_dict = {'a': np.arange(2),
     'b': np.arange(2)}
m_mydict = MyUserDict(m_dict)

Then I meet a problem. When turning a list of UserDict into numpy array, values in UserDict have lost:

m_dict_array = np.array([m_dict, m_dict])
m_mydict_array = np.array([m_mydict, m_mydict])
print(m_dict_array)
print(m_mydict_array)

""" The output is
[{'a': array([0, 1]), 'b': array([0, 1])}
 {'a': array([0, 1]), 'b': array([0, 1])}]
[['a' 'b']
 ['a' 'b']]
"""

The solution I find is adding __array__() method to MyUserDict:

class MyUserDict(UserDict):
     def __array__(self):
        # Prevent NumPy from converting to an array of keys
        return np.array(self.data, dtype=object)

m_mydict = MyUserDict(m_dict)
m_mydict_array = np.array([m_mydict, m_mydict])
print(m_mydict_array)

This indeed solve my problem. But another problem I quickly realized is that self.data is plain dict and elements in m_mydict_array may be also plain dict. So, I check the type as follows:

print(type(m_mydict_array[0]))

The type is MyUserDict I defined. This is certainly the result I hoped for, but I have no idea why it turned out like this.

How does __array__() work? Why is the type of m_mydict_array[0] not default 'dict'?

dtype=object means (roughly) to use plain Python objects without any conversion. In fact, m_mydict_array[0] is m_mydict is true. — ken
– ken, Commented Jun 10 at 8:18
The interesting point to this question is that with np.array([m_mydict, m_mydict]) the __array__ method is actually invoked, however the resulting array does not contain the return value of __array__ but rather the object itself. Changing the return to a dummy array (e.g. np.arange) does in fact change the result completely. — Daraan
– Daraan, Commented Jun 10 at 9:56

Daraan · Accepted Answer · 2025-06-10 10:53:40Z

0

I am not entirely sure if this is bug, but it is unexpected. I opened this Issue on GitHub.

I workaround & repo the issue in a different way so it is more obvious to readers. Replace None with self.data in your case.

The bug can be reproduced with:

import numpy as np

class MyUserDict:  # UserDict not necessary.
    def __array__(self, dtype=None, copy=None):
        return np.array(None, dtype=object)

arr = np.array([MyUserDict()])
print(arr)  # [{1:1}]
print(type(arr[0])) 
Output
[<__main__.MyUserDict object at 0xXX]
<class '__main__.MyUserDict'>

Expected:
[None]
NoneType

You receive the correct value when you do not return a single item but with at lease one dimension:

    def __array__(self):
        return np.array(None, dtype=object, ndim=1)

    def __array__(self):
        return np.array([None], dtype=object)

answered Jun 10 at 10:53

Daraan

5,1267 gold badges24 silver badges51 bronze badges

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

1 Comment

DrizzleatDusk Jun 11 at 0:47

Thanks for your enthusiastic reply and I have noticed explanations on Github from 'seberg'. But I still do not get the point😂. I don't understand how the explanation is related to the behavior of the __array__() method.

Collectives™ on Stack Overflow

Numpy array not working when return array with single item

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related