fix: an unexpected 'equal_nan' keyword argument #66
Conversation
…nt 'equal_nan' 'equal_nan' was introduced from Numpy 1.24 https://numpy.org/doc/1.24/reference/generated/numpy.unique.html
There was a problem hiding this comment.
@jpivarski - please, check this PR when you have time. It is needed for @ohrechykha's PR #57 to pass. Thanks!
There was a problem hiding this comment.
Without the equal_nan keyword argument, we can't satisfy the protocol specified by the Array API (without doing extra work with np.isnan to collapse multiple NaN values into a single NaN), and this was added in NumPy 1.24.
It's fine for ragged to have a tighter NumPy requirement than Awkward Array. (We don't want to increase Awkward's NumPy requirement without deliberation, since it will constrain the set of packages that can be installed in an environment.)
NumPy 1.24 is the oldest currently-supported version of NumPy, and even then, it will be discontinued this December.
For these reasons, I think I can merge it. Does ragged have a requirements-min.txt or similar that also needs to be updated? No, I don't see one, and NumPy is not explicitly mentioned in .readthedocs, either. Oh, but if we increase the minimum tested NumPy version in .github/workflows/ci.yml, we must also increase the actual installation constraint for PyPI. I'll do that now, and then merge after tests pass.
There was a problem hiding this comment.
@jpivarski - the PR cannot be auto merged because it is waiting for a required 1.22 workflow to run. I don't think, I have rights to disable it.
|
Got it! |
Fixes TypeError: _unique_dispatcher() got an unexpected keyword argument 'equal_nan'
'equal_nan' was introduced from Numpy 1.24
https://numpy.org/doc/1.24/reference/generated/numpy.unique.html