In the NumPy documentation, it is demonstrated that you can add a custom attribute by subclassing the ndarray:

import numpy as np

class RealisticInfoArray(np.ndarray):
    def __new__(cls, input_array, info=None):
        obj = np.asarray(input_array).view(cls)
        obj.info = info
        return obj

    def __array_finalize__(self, obj):
        if obj is None: return
        self.info = getattr(obj, 'info', None)

arr = np.arange(5)
obj = RealisticInfoArray(arr, info='information')
print(obj.info)

This example works perfectly fine. However, if I want to make the new attribute “private” (i.e., by calling the attribute self._info):

import numpy as np

class RealisticInfoArray(np.ndarray):
    def __new__(cls, input_array, info=None):
        obj = np.asarray(input_array).view(cls)
        obj.info = info
        return obj

    def __array_finalize__(self, obj):
        if obj is None: return
        self._info = getattr(obj, 'info', None)

arr = np.arange(5)
obj = RealisticInfoArray(arr, info='information')
print(obj._info)

Then None is printed. Can somebody help me understand why this is and how to fix this so that obj._info is not None and behaves the same as the first example?