✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 133d82a. Link to the linter CI: here}

Scipy quantile now supports array API (but not weights): scipy/scipy#22352

Their quantile function is probably (?) faster than ours (benchmarking needed) and it would be nice to not maintain another array API function. But it was added after our min scipy.

I'll open an RFC issue about adding quantile to array api extra, it's been on my todo for a while.

data-apis/array-api-extra#340 may take a long while, so we should still decide here what we want to do, even if it is 'temporary'.

Thanks @lucyleeow!

Do we really need a dedicated implementation for quantile in array-api-extra when all the relevant array api libraries support it (I checked that dask and jax also support this aside from cupy and Pytorch)? Maybe all we actually need is a helper that can call the respective function of the library under consideration (Pytorch has a different name for the dim parameter) and simply convert to numpy to handle array-api-strict.

Do we really need a dedicated implementation for quantile in array-api-extra when all the relevant array api libraries support it

1. The dedicated implementation should not be too difficult, as https://github.com/scipy/scipy/blob/main/scipy/stats/_quantile.py already supports the standard

2. The delegation to existing libraries you describe can be covered by array-api-extra too: https://github.com/data-apis/array-api-extra/blob/main/src/array_api_extra/_delegation.py

2. The delegation to existing libraries you describe can be covered by array-api-extra too: https://github.com/data-apis/array-api-extra/blob/main/src/array_api_extra/_delegation.py

@lucascolley Thanks for the feedback. I think it makes sense to add it there then. Do we need a new implementation though when we don't really need it? I guess it would require copying the relevant code from scipy?

Do we need a new implementation though when we don't really need it?

Yes, it is good to future-proof and not rely on existing libraries. cf. Tim's comment over at data-apis/array-api-extra#340 (comment) let's continue discussion over there