NumPy 2.0 compatibility

[qiyunzhu]

It seems that scikit-bio is compatible with the upcoming NumPy 2.0. Following this guideline, I ran:

ruff check skbio --select NPY201

And nothing happened, meaning the check passed. However, I haven't been able to actually install NumPy 2.0 for test (probably because it is still under active development). Therefore we should keep an eye on it.

Thanks @wasade for pointing out this.

[rgommers]

You can install numpy 2.0.0rc1 from PyPI now, with pip install numpy --pre. It'd be great to get a confirmation that everything works well with numpy 2.x

[qiyunzhu]

@rgommers Thanks for pointing us to this!

Just tested. Got error:

File "h5py/_errors.pyx", line 1, in init h5py._errors
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

This is probably an issue in h5py, a dependency of scikit-bio. I saw this discussion in the h5py codebase. We probably need to wait until after they release a NumPy 2.0 compatible version. @wasade

[rgommers]

Thanks for testing @qiyunzhu. Good to know about h5py, I've asked on the issue you linked.

[qiyunzhu]

@rgommers Update: I was able to install numpy 2.0.0rc1 and compatible versions of scipy, pandas and h5py. Then got the following error when importing scikit-bio:

  File "skbio/stats/__subsample.pyx", line 1, in init skbio.stats.__subsample
ImportError: numpy.core.multiarray failed to import (auto-generated because you didn't call 'numpy.import_array()' after cimporting numpy; use '<void>numpy._import_array' to disable if you are certain you don't need it).

We have planned to replace the __subsample function with a new one so this error might become irrelevant in the next release. But if you have some insights to share we will appreciate.

[rgommers]

Not too much more than what the error message says:

because you didn't call 'numpy.import_array()' after cimporting numpy

in the Cython file where this error comes from, I assume there's a cimport numpy as cnp statement. and no cnp.import_array().

[rgommers]

h5py release is out now - slowly getting there:)

[qiyunzhu]

Thanks for the update! Now we are waiting for Pandas 2.2.2. Then we add numpy 2.0 support for the upstream package biom-format. Then we can unblock numpy 2.0 in scikit-bio.

[rgommers]

Now we are waiting for Pandas 2.2.2.

That's up on PyPI too now, since ~25 minutes ago.

[rgommers]

Hi @qiyunzhu, I just wanted to check in on this, and ask if you plan to include support in 0.6.1 or 0.7.0 (and, if you know, roughly when)?

[qiyunzhu]

Hello @rgommers , We recently resolved the NumPy 2.0 support issue in the upstream package biom-format (see here). We are about to resolve this in scikit-bio in the next few days. It will be 0.6.1. Let's say it will be released in a month or so.

[rgommers]

Thanks! I'm not 100% sure that the final numpy 2.0 release will take more than a month from now, so a release before the end of May would be safer. But if you've got all the problems resolved then that's already good to know.