Evgeni Burovski
Evgeni Burovski
> result is a dask array, type(result) is dask.array.Array Then the fix is to wrap the `# lists/tuples` stanza in an explicit `if isinstance(result, (list, tuple) `
Looks like indeed some np.asarray / np.asanyarray calls are missing and internal functions receive mixtures of numpy and dask arrays: ``` 917 -> def _select(input, labels=None, index=None, find_min=False, find_max=False, 918...
The reported result is smaller than the float64 machine epsilon. The OP example involves casting between integers (via `np.datetime[ns]`, float32 and float64). The integers themselves are large, and it's not...
Thank you @rkern !
Thanks @jameswilburlewis for the clarification! I'm going to close this issue then, in favor of https://github.com/pydata/xarray/issues/9404. Can continue the discussion here too, of course.
git blame points to https://github.com/scipy/scipy/pull/5902 as the PR which changed from delegating to `np.interp` for all float dtypes to float64 only. It does not seem to mention float32 specifically, but...
Looks like all review feedback was addressed. To keep the ball rolling, I think it makes sense to land this in about a week's time --- unless there are further...
@j-bowhay Your review is listed as _changes requested_, so you're looking for some more time to review this, correct? No pressure, am just trying to avoid this PR falling through...
Nah, next week sounds great :-) . And good luck with the thesis!
> The general xp_assert_close(np_array, np_array) must always pass, regardless of check_0d or not. +1