Stephan Hoyer

> This API has particular value for complex numbers, as the alternative `abs(x)**2` requires taking the square root and the alternative `x*conj(x)` potentially requires additional memory allocation before computation. I...

Two thoughts: 1. We could make `axis` optional only in the case of 1D arrays 2. I would not necessarily be limited by constraints on `np.put`, which in my opinion...

I think 3-way True/False/None is a totally reasonable solution if we are designing this API from scratch. The problem is that right now `copy=False` means "copy if needed" for NumPy,...

Yes, I was advocating for enums last time 🤦 I commented on the mailing list discussion advocating for True/False/None.

I think it's equally straightforward to write an explicitly stateful RNG system like numpy.random.Generator using JAX. Here's a prototype: ```python import jax class JaxGenerator: def __init__(self, state): self.state = state...

To bridge the gap between NumPy's and JAX's random number APIs, I would suggest slightly extending NumPy's API so it's easier to be explicit about state. Namely, we should add...

I wonder if we could at least extend the Array API standard to allow for arrays with novel dtypes for storage but not computation. For example, you could convert a...

> I don't care too much about this, but dask.Array people may know better if it is useful for them. Promotion was explicitly mentioned also in https://github.com/data-apis/array-api/issues/403. For the most...

> However for libraries that have arrays with unknown shape elements (like Dask). `nonzero` can run into some difficulties as the different arrays may not be `concatenate`d (at least not...

Right, I guess my question is why getting all indices together into a single array matters.