[BUG] Indexing inconsistencies

[dionhaefner]

Describe the bug

• mlx.core.newaxis is missing:

>>> mx.array([1])[mx.newaxis]
AttributeError: module 'mlx.core' has no attribute 'newaxis'

• Cannot index with Ellipsis / ...:

>>> mx.array([1])[...]
ValueError: Cannot index mlx array using the given type.

To Reproduce

See above.

Expected behavior

Consistency with NumPy.

Desktop (please complete the following information): v0.0.7

[awni]

Cool, I don't mind adding that though I find it a bit odd how numpy has different syntax for the same thing.

In mx we went with None for adding new axes (which numpy aslo has). So you can do:

mx.array([1])[None]`

I think for this we could bind a little axis type in pybind11 and then use it in the same place we us None.

[angeloskath]

No need. np.newaxis is an alias for None, meaning that np.newaxis is None returns True. All we have to do is add an alias as mx.newaxis = None if we want full numpy compatibility.

[AvikantSrivastava]

@awni @angeloskath Hi, can I pick this up?

[awni]

Yes sure!

[AvikantSrivastava]

@awni should I add all the constant that numpy offers? https://numpy.org/doc/stable/reference/constants.html

I was thinking of adding a init_constants(m) in mlx/python/src/mlx.cpp

[AvikantSrivastava]

@awni @angeloskath fixed in https://github.com/ml-explore/mlx/pull/428

can you folks take a look at the PR

[awni]

Sent a comment, but o/w looks great, thanks for adding thos!