Pyomo is not compatible with Numpy 2.0.0

[JeanWi]

Summary

Pyomo is not compatible with numpy 2.0.0 as np.float_ was removed.
From the numpy release note: Alias np.float_ has been removed. Use np.float64 instead. (https://github.com/numpy/numpy/releases/tag/v2.0.0)

Steps to reproduce the issue

Installing the latest numpy (2.0.0) and using pyomo leads to the following error:
AttributeError: np.float_ was removed in the NumPy 2.0 release. Use np.float64 instead.. Did you mean: 'float16'?

The problem is caused in: pyomo\common\dependencies.py:1002

Error Message

C:\hostedtoolcache\windows\Python\3.12.3\x64\Lib\site-packages\pyomo\common\dependencies.py:475: in exec_module
deferred.resolve()
C:\hostedtoolcache\windows\Python\3.12.3\x64\Lib\site-packages\pyomo\common\dependencies.py:334: in resolve
self._module, self._available = _perform_import(
C:\hostedtoolcache\windows\Python\3.12.3\x64\Lib\site-packages\pyomo\common\dependencies.py:768: in _perform_import
callback(module, True)
C:\hostedtoolcache\windows\Python\3.12.3\x64\Lib\site-packages\pyomo\common\dependencies.py:1002: in finalize_numpy
floats = [np.float, np.float16, np.float32, np.float64]
C:\hostedtoolcache\windows\Python\3.12.3\x64\Lib\site-packages\numpy_init.py:411: in getattr
raise AttributeError(
E AttributeError: np.float_ was removed in the NumPy 2.0 release. Use np.float64 instead.. Did you mean: 'float16'?

### Information on your system

Pyomo version: 6.7.3
Python version: 3.12
Operating system: Windows 11
How Pyomo was installed (PyPI, conda, source): PyPI
Solver (if applicable): gurobi, but not applicable


### Additional information

None

[jasonmbray-p66]

Thanks for working on this!

My conda environment just broke yesterday when I rebuilt my environment and it picked up numpy 2 automatically. However, I'm working from a pinned version of pyomo due to conflicts with other packages in my project and won't be able to take advantage of this fix in the latest pyomo version when it is released. I can pin numpy manually in my requirements, but to help others avoid the same issue, is it possible to also update the dependency version requirements of past pyomo versions so that conda, pip, and other package managers know that previous versions of pyomo require numpy<2?

[mrmundt]

Hi, @jasonmbray-p66 - A couple of notes:

1. Some of our offerings still won't be numpy>=2.0.0 compatible for a while. For example, the solver xpress is not compatible. Our efforts are to make base Pyomo compatible - but can't guarantee widespread compatibility!
2. We did pin numpy<2.0.0 for the two most recent releases (6.7.3, 6.7.2). We will discuss doing so for previous versions at our next dev team call.

[jasonmbray-p66]

Just my luck that my current project is pinned to <= 6.7.1, so I just barely missed the updated pins! Thanks for considering my request.

[p-snft]

Hi, @jasonmbray-p66 - A couple of notes:
2. We did pin numpy<2.0.0 for the two most recent releases (6.7.3, 6.7.2). We will discuss doing so for previous versions at our next dev team call.

I just installed the latest version of Pyomo and it pulled numpy 2.0 (Numpy without any pinned version was needed a package that also needs .) So, pinning did not seem to work.

I have a theory about this. Citing setup.py of the most recent release, you find (a lot of lines omitted)

setup_kwargs = dict(
    python_requires='>=3.8',
    install_requires=['ply'],
    extras_require={
        'optional': [
            'numpy<2.0.0',
        ],
    },
)

that it declares "numpy<2.0.0" as an optional dependency. So, if another package has install_requires=['numpy'] listed as a mandatory requirement, will it still install "numpy>=2.0.0" because the optional dependency of pyomo is just ignored? At the moment, I assume so. (An example showing this in can be found in the oemof.solph test pipeline at https://github.com/oemof/oemof-solph/actions/runs/9577740874/job/26406541517#step:6:23. Note that this pipeline is exclusively using pip for package management, so unlike watertap, it cannot be because of conda.)

[mrmundt]

Note for all tracking this:

• We will not be pinning numpy<2.0.0 for releases older than <6.7.2 - we do not have the capacity to re-release all of the old versions
• numpy has always been an optional dependency and will continue to be so. This is causing the issue noted by @p-snft and is intended behavior.

[dagoberto-escobar]

I got the same issue while using Numpy v2.0.1 along with the most recent release v6.7.3, and got to circumvent the issue by adding the lines below

import numpy as np
np.float_ = np.float64
np.complex_=np.complex128

before importing Pyomo

[mrmundt]

The next release (6.7.4) will have this issue resolved.