hash(ndindex_obj) != hash(raw_obj) about ndindex HOT 5 CLOSED

We also have

>>> ndindex(False) == ndindex(0)
False
>>> False == 0
True

because

>>> ndindex(False).raw
array(False)

Not to mention this behavior

>>> ndindex([0, 1]) == np.array([0, 1])
True
>>> np.array([0, 1]) == ndindex([0, 1])
array([False, False])

which we can't do anything about. The best recommendation is to always convert any index object into an ndindex object, and manipulate that, and only use a raw index type (.raw) at the end when you actually index an array.

from ndindex.

Fix at #93

from ndindex.

Not to mention this behavior

>>> ndindex([0, 1]) == np.array([0, 1])
True
>>> np.array([0, 1]) == ndindex([0, 1])
array([False, False])

@asmeurer In your ideal world, which of the following would be the behavior for ndarray vs NDIndex equality?

A)

ndindex([0, 1]) == np.array([0, 1])
> array([True, True])
np.array([0, 1]) == ndindex([0, 1])
> array([True, True])

or B)

ndindex([0, 1]) == np.array([0, 1])
> True
np.array([0, 1]) == ndindex([0, 1])
> True

Both can be done. I'd argue for A), since that's now the expected behavior for "array-like" objects (as defined by numpy as of 1.17, I think?). B) is also doable, but it will have to have the side-effect that isinstance(ndindex([0, 1]), np.ndarray) is True

from ndindex.

I would prefer it if I could make them both give True. == giving an array makes very basic things like checking if an index is in a list or dictionary difficult, because bool() fails on arrays. And anyway, an ndindex array index object is not an array-like. See https://quansight.github.io/ndindex/type-confusion.html#integerarray-and-booleanarray

from ndindex.

I opened a new issue for this over at #95. Let's continue the conversation there

from ndindex.