Add array conversion methods to ndarray

[sproshev]

Actual types are like:

def item(self, *args: int) -> T: ...
@overload
def itemset(self, __value: T): ...
@overload
def itemset(self, __item: _ShapeLike, __value: T): ...
def astype(self,
           dtype: _DtypeLike,
           order: str=...,
           casting: str=...,
           subok: bool=...,
           copy: bool=...) -> ndarray[dtype]: ...
def byteswap(self, inplace: bool=...) -> ndarray[T]: ...
def copy(self, order: str=...) -> ndarray[T]: ...
def view(self, dtype: Union[_DtypeLike, NdArrayOrMatrix]=..., type: NdArrayOrMatrix=...) -> NdArrayOrMatrix[dtype]: ...
def getfield(self, dtype: Union[_DtypeLike, str], offset: int=...) -> ndarray[dtype]: ...

[shoyer]

Thanks for these PRs! I'll just point out that we also have tests setup that might be worth augmenting for any complex cases: https://github.com/numpy/numpy-stubs/tree/master/tests

[shoyer]

This should probably substitute Type -> Type[ndarray] and Any -> Type[ndarray].

[sproshev]

Thank you! Fixed. Covered with tests

[shoyer]

This is great, but just to be clear you don't need tests for everything -- especially not for simple methods like dumps() that just returns bytes.

I would save tests for complex cases where it's not immediately evident that the code is correct, e.g., where you have overloads or type variables.

[shoyer]

In theory, I think this should be numpy.matrix.

Or we can could make a dummy subclass, if np.matrix isn't defined, e.g.,

class SubArray(np.ndarray):
    pass

reveal_type(nd.view(type=SubArray))  # E: SubArray

Does that actually work?

[sproshev]

Thank you for review! I've removed simple tests and fixed type hints for view.