Athan
As discussed in the Sept 22, 2022, consortium meeting, the plan is to merge this and https://github.com/data-apis/array-api/pull/446 as is. We can revisit the decision to only support complex dtypes based...
The broadcasting behavior of empty arrays according to the spec is as intended. Namely, in general, when an operand is an empty array, one should get an empty array result....
@oleksandr-pavlyk Correct. They are not consistency poisonous in the example you use as, if size is > 1 and the dimensions do not match, the arrays do not broadcast regardless...
Given that size `0` dimensions could introduce confusion, we should probably add examples demonstrating this in the broadcasting specification. See https://github.com/data-apis/array-api/pull/407.
@leofang Re: defn branch cut. I can move to the list of specification definitions. Re: same branch cuts. Currently, I am approaching this on a case-by-case basis. For `sqrt`, the...
Re; broadcasting. My sense is that it would be best for PyTorch to consider its current behavior a "bug" and align with NumPy et al in terms of broadcasting. As...
Corresponded with @Lezcano and the plan is for PyTorch to patch its current broadcast behavior in `linalg.cross` to match NumPy's `cross` behavior. Namely, - the dimension over which cross product...
This was addressed in https://github.com/data-apis/array-api/pull/417. Closing.
Can probably simplify the example here using arrays which are explicitly not of the same rank: ```python In [15]: x = np.zeros((3,4)) In [16]: x Out[16]: array([[0., 0., 0., 0.],...
Proposal: add APIs for getting and setting elements via a list of indices (i.e., `take`, `put`, etc)
@rgommers Thanks for the comments. 1. Correct. I've updated the table with Torch and TF scatter and gather methods. 2. Correct me if I am wrong, but `indices.size` need not...