Updating to match Numpy 2.0 requirements

[ericspod]

Fixes #7856.

Description

This introduces changes to meet Numpy 2.0 requirements. MONAI itself is compatible with Numpy 2.0 however some dependencies are not such as older versions of Pytorch. This PR adjusts the MAX_SEED value to be compatible with Numpy 2.0 behaviour changes, uses the ptp function, and some other minor tweaks. The versions for dependencies are also fixed to exclude Numpy 2.0.

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[Nic-Ma]

Hi @ericspod ,

Thanks for your quick update for Numpy 2.0.
May I know whether you found any test failures due to 3rd party dependency can't support Numpy 2.0 so far?

Thanks in advance.

[ericspod]

These appear to be NOT compatible with Numpy2:

• Pytorch<2.3
• Pytorch 2.4.0+cpu on Windows?
• Pyarrow
• ClearML when using Pyarrow

[ericspod]

@KumoLiu We'll leave the requirements excluding Numpy2 but otherwise updating the minimum version.

[KumoLiu]

/build

[KumoLiu]

Thanks for the PR, looks good to me.

[michael-camilleri]

Will Numpy 2 be supported soon? Especially as the dependency is due to older versions of pytorch/other libraries and not explicitly support within MONAI itself? is there a reason why MONAI does not drop older versions of pytorch?

[ericspod]

Hi @michael-camilleri The dependency issue for us is that we do want to continue supporting older versions of Pytorch because many users may need older versions for compatibility or because their environment requires it. The problem that everyone has discovered is that their versions strings for Numpy would set a minimum requirement but not a max, so Numpy 2.0 with its incompatibility isn't excluded. Older versions of libraries are unlikely to be updated with support for Numpy 2.0, and we have no way of forcing Numpy 1.x when these are present. For the moment we've chosen to just restrict the version although MONAI itself should be compatible now with the changes made. I think it would work for you to try installing MONAI in an environment (venv or conda) and then installing Numpy 2.0 subsequently.

[michael-camilleri]

Thanks.
Indeed, that is what I did and seems to be working. Wondering if there is a way of supporting this versioning in the documentation... I did not realise that numpy had been downgraded until a colleague (who needed 2.0) came across some code errors.

[ericspod]

I agree, we can improve the documentation and mention this issue with Numpy 2. Beyond that, we would need a way of directly controlling version resolution in pip to account for known incompatibilities, but this isn't currently possible and would rely on guessing which libraries have a 2.0 problem.

[michael-camilleri]

Yes, I see your point. I think a documentation of the issue within the installation instructions is enough to be honest, given that it is (in most cases) an easy fix. Michael

…

On Wed, Nov 27, 2024 at 5:04 PM Eric Kerfoot ***@***.***> wrote: I agree, we can improve the documentation and mention this issue with Numpy 2. Beyond that, we would need a way of directly controlling version resolution in pip to account for known incompatibilities, but this isn't currently possible and would rely on guessing which libraries have a 2.0 problem. — Reply to this email directly, view it on GitHub <#7857 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIKRP37QOIVKWWXZVV7OQDL2CX3QPAVCNFSM6AAAAABJOC2FRGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMBUGM3TQOBSGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>