[Neuroimaging] ANN: NiBabel 5.0

[Christopher Markiewicz]

Hi all,

I'm pleased to announce the release of NiBabel 5.0!

This version requires Python 3.8 and numpy 1.19, in accordance with our version support schedule <https://github.com/nipy/nibabel/issues/803>.

As with other major version releases, there are DeprecationWarnings that will now become ExpiredDeprecationErrors, and ExpiredDeprecationErrors that will now become AttributeErrors, as the deprecated features are progressively removed.

We had a shortened 4.x lifetime because it was discovered that a growing number of images were being created with int64 data types and causing problems with other neuroimaging tools. While the NIfTI standards permit int64 data, it is not widely adopted and there is no clear use case that is not better handled by float64. Therefore in 4.0, we introduced a warning if users generated int64 images (easy to do with `Nifti1Image(data.astype(int), affine)`) without providing an explicit dtype (or pre-generated header). With 5.0, this warning has become an error.

Additional changes in 5.0 include an updated SerializableImage interface that allows you to serialize images to/from custom filehandles with to_stream()/from_stream(), as well as load images from URLs using, e.g., Nifti1Image.from_url().

With many thanks to everybody who reported issues, submitted patches, and especially reviewed PRs.

Please test out the new version and report any issues you have to https://github.com/nipy/nibabel/issues.

The full changelog can be found at https://github.com/nipy/nibabel/releases/tag/5.0.0.

Best,
-- 
Christopher J. Markiewicz, PhD
Center for Reproducible Neuroscience
Stanford University