numpy errors

[dineshbvadhia]

The mypy.ini file includes:

plugins = numpy.typing.mypy_plugin

On python+numpy code, mypy throws errors of the form:

error: Incompatible types in assignment (expression has type "Union[Any, _SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]", variable has type "int")

For example, the line:

indices: np.ndarray = this_vector.nonzero()[0]

causes the above error. If changed to a list:

indices: list[float] = this_vector.nonzero()[0].tolist()

it works but defeats the purpose of using numpy which performers faster and the code is cleaner.

Please explain what is going on. Thanks

[BvB93]

Can you provide an actual fully self-contained example?
The error message suggests that you're using npt.ArrayLike for an unintended purpose, but with the currently provided code snippets I can't make out anything more conclusive.

[dineshbvadhia]

Sure:

import numpy as np
import numpy.typing as npt

def get_indices(v: npt.ArrayLike) -> np.ndarray:
    idx: np.ndarray = v.nonzero()[0]
    return idx

def main():
    vec: np.ndarray = np.asarray([1.0, 0, 5.0, 7.0, 0], dtype=np.float32)
    print('vector:', type(vec), vec)

    indices: np.ndarray = get_indices(vec)
    print('indices:', type(indices), indices)

if __name__ == '__main__':
    main()

> python bad_np.py
vector: <class 'numpy.ndarray'> [1. 0. 5. 7. 0.]
indices: <class 'numpy.ndarray'> [0 2 3]

> mypy bad_np.py --no-incremental
bad_np.py:5: error: Item "_SupportsArray[dtype[Any]]" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "_NestedSequence[_SupportsArray[dtype[Any]]]" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "bool" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "int" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "float" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "complex" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "str" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "bytes" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
bad_np.py:5: error: Item "_NestedSequence[Union[bool, int, float, complex, str, bytes]]" of "Union[_SupportsArray[dtype[Any]], _NestedSequence[_SupportsArray[dtype[Any]]], bool, int, float, complex, str, bytes, _NestedSequence[Union[bool, int, float, complex, str, bytes]]]" has no attribute "nonzero"
Found 9 errors in 1 file (checked 1 source file)

[BvB93]

def get_indices(v: npt.ArrayLike) -> np.ndarray:
    idx: np.ndarray = v.nonzero()[0]
    return idx

This explains a lot. npt.ArrayLike is a union of all types that can safely(-ish) be converted into an np.ndarray, including things like scalars and nested lists; it does not represent some type of duck-array.

The error is correct in this case, what you'd want here is to either call the np.nonzero function (not the method) or to swap the npt.ArrayLike annotation with np.ndarray.

[dineshbvadhia]

If the array type is known (eg. int, float) then use np.ndarray as the annotation instead of npt.ArrayLike - is that correct?

[BvB93]

If the array type is known (eg. int, float)

Yes, assuming you're talking about the arrays' dtype here.
In the current state of your example you're implying that the likes of get_indices([0, 1, 2, 3]) should be legal though via the use of npt.ArrayLike.