Jake Vanderplas

> * Both JAX and TensorFlow allow `x` to be multi-dimensional I don't think this is correct in the case of JAX. I confirmed that this errors in v0.5.x and...

> @jakevdp Would be good to update the docstring then for `bincount` Thanks for pointing that out – updated in https://github.com/jax-ml/jax/pull/29441.

Thanks for the fix! Could you also add a regression test for this new behavior? @tlu7, could you take a look?

So I'm not totally familiar with this code: could you explain the fix and why the updated test addresses the problem?

What about booleans? The Array API spec doesn't mention these here, but it seems like it would be consistent to return boolean outputs for boolean inputs to these functions.

Can you provide a short reproducible example of code that produces invalid output?

Altair relies on Python's built-in `json` module to serialize user input to JSON. It seems the bug is in the Python standard library: ```python >>> import json >>>json.dumps({'x': float('nan')}) {"x":...

Ah, maybe we should be using something like ``allow_nan=False``: ```python >>> json.dumps({'x': float('nan')}, allow_nan=False) --------------------------------------------------------------------------- ValueError Traceback (most recent call last) in () 1 import json ----> 2 json.dumps({'x': float('nan')},...

Hi all – we just released `ml_dtypes`: https://github.com/jax-ml/ml_dtypes. It's a lightweight stand-alone implementation of bfloat16 and a few other dtypes used in machine learning contexts, extracted from the source code...

Switch is implemented in terms of `cond`, so yes it has the same characteristics.