Jake Vanderplas

Thanks for the report! This is definitely unexpected, and points to some compiler issue. I updated your timing to separate out the first call, use `block_until_ready` to avoid issues due...

Thanks! This will have to wait for a vega-datasets release that includes the penguins data.

Thanks for the report! in general `reduce_max` over a sparse structure is essentially equivalent to a group-by reduction over the data buffer with respect to the index buffer, and there's...

Please remove the unused import & fix conflicts! Also, please comment when you've done this because I don't get notifications if you just push the change (that's why I'm so...

One note: `SupportsDType` is not entirely accurate, because there is at least one example of a type that matches its protocol and yet is not compatible as an argument to...

> I'd rather not use Protocols. To be close to accurate in describing valid inputs of `numpy.dtype`, you need protocols. This is a NumPy thing, not a JAX thing (`jnp.dtype`...

The PR description notes backward compatibility, but it should be updated to make clear that although the API is backward compatible, the outputs are not: **the function outputs a different...

I also think it would be worth discussing whether this breakage is advisable. I'd lean toward `fit_mean=False` by default (at least for now) to avoid this kind of breakage, despite...

> JAX has interpreted `default_device()` as "whatever the `device=` kwarg will accept", and returns None. We were doing our best to interpret https://data-apis.org/array-api/2023.12/design_topics/device_support.html in a way that was compatible with...

> I'm reading the JAX documentation and it gives me the impression that when you call a creation function (`empty`, `asarray`, `ones`, etc.) the array _will_ be created on whatever...