From numpy documentation for np.double it says np.float and np.float64 are aliases for np.double. (Noting the docs say Alias on this platform for np.float64)

I was expecting mypy to treat the aliases similarly, but it didn’t.

Here is an example (print_max.py):

import numpy as np
from numpy.typing import NDArray

def print_max(arr: NDArray[np.float32]) -> None:

    print(f"arr.max() = {arr.max()}")

a = np.ones((2,3), dtype=np.float_)
b = np.ones((2,3), dtype=np.float64)
c = np.ones((2,3), dtype=np.double)
d = np.ones((2,3), dtype=np.float32)

print_max(a)
print_max(b)
print_max(c)
print_max(d)

If I run mypy print_max.py, I get the following:

practice.py:14: error: Argument 1 to "print_max" has incompatible type "ndarray[Any, dtype[floating[_64Bit]]]"; expected "ndarray[Any, dtype[floating[_32Bit]]]"  [arg-type]
Found 1 error in 1 file (checked 1 source file)

I would have expected mypy to have found errors for cases when the ones array was defined with with dtypes np.float_, np.float64, & np.double since the type hint in print_max specified np.float32.

Why is numpy treating these aliases for 64-bit floating points (np.float_, np.float64, & np.double) differently?