Python bool与numpy bool_的行为差异及相关技术咨询

Great catch on the quirky behavior between numpy's bool_ and Python's native bool! Let's tackle your questions one by one.

1. General Behavior Differences Beyond Numpy-Specific Methods #

While numpy.bool_ is built to be compatible with Python's bool, there are several subtle but impactful differences in their everyday behavior:

• Identity checks (is) fail: Python’s True and False are singletons—there’s only one instance of each in the runtime. Numpy creates new bool_ instances every time you initialize one, so np.bool_(True) is True will always return False, even though their values are equivalent.
• Type distinctions: type(np.bool_(True)) returns <class 'numpy.bool_'>, while type(True) returns <class 'bool'>. That said, isinstance(np.bool_(True), bool) will still return True because numpy’s scalar types are registered as subclasses of their Python counterparts for compatibility.
• Negation behavior differs: Python’s bool uses not for logical negation, and ~ performs bitwise negation (e.g., ~True returns -2 since True maps to integer 1). For numpy.bool_, ~np.bool_(True) correctly returns False—numpy overrides the bitwise operator to act as logical negation for boolean scalars.
• Operation type persistence: Arithmetic operations on numpy.bool_ keep the result as a numpy numeric type. For example:

  np.bool_(True) + 1  # Returns 2 as numpy.int64
  True + 1            # Returns 2 as native int
  

• Hashing & serialization differences: The hash value of a numpy.bool_ instance doesn’t match its Python bool equivalent, which can cause issues if you use them as dictionary keys alongside native booleans. Tools like pickle also treat them as distinct types during serialization.

2. How to Check if a Numpy Array Contains Only numpy.bool_ #

The simplest and most efficient method is to verify the array’s dtype:

import numpy as np

# Numpy bool_ array
arr_bool = np.array([1, 0, 2, 0], dtype=bool)
print(arr_bool.dtype == np.bool_)  # Returns True

# Object array with Python bools
arr_obj_bool = np.array([True, False, True, False], dtype=object)
print(arr_obj_bool.dtype == np.bool_)  # Returns False

If you need to handle edge cases where an object-dtype array might mix numpy.bool_ and Python bool (a rare scenario), you can validate each element (note: this is slower for large arrays):

def is_all_numpy_bool(arr):
    return all(isinstance(elem, np.bool_) for elem in arr.flat)

# Test mixed array
mixed_arr = np.array([np.bool_(True), False, np.bool_(False)], dtype=object)
print(is_all_numpy_bool(mixed_arr))  # Returns False

For a more robust dtype check (in case of alias types), use np.issubdtype:

print(np.issubdtype(arr_bool.dtype, np.bool_))  # Returns True

Quick Reminder #

As you noted, using is to compare boolean values is bad practice overall. Always prefer direct boolean checks like if greeting: instead of if greeting == True: or if greeting is True—this works seamlessly for both Python bool and numpy.bool_.

内容的提问来源于stack exchange，提问作者vurmux