[BUG] Dependencies version checking not handling python release candidate correctly

[zobelb]

Describe the bug

When using Python 3.11.0rc1, an exception is thrown stating this is not a release less than 3.12. I suspect that there being no space between 3.11.0 and rc1 is causing this version test to fail.

An exception occurred: IndividualInceptionClassifier requires python version to be <3.12, but system python version is 3.11.0rc1 (main, Aug 12 2022, 10:02:14) [GCC 11.2.0].

Steps/Code to reproduce the bug

from aeon.datasets import load_unit_test

X_train, y_train = load_unit_test(split="train")
X_test, y_test = load_unit_test(split="test")

inctime = InceptionTimeClassifier(n_epochs=20,batch_size=4)
print(inctime.fit(X_train, y_train))

Expected results

InceptionTimeClassifier(batch_size=4, n_epochs=20)

Actual results

File "/home/username/aeon_test.py", line 10, in <module>
    print(inctime.fit(X_train, y_train))
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/aeon/classification/base.py", line 175, in fit
    self._fit(X, y)
  File "/usr/local/lib/python3.11/dist-packages/aeon/classification/deep_learning/_inception_time.py", line 249, in _fit
    cls = IndividualInceptionClassifier(
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/aeon/classification/deep_learning/_inception_time.py", line 540, in __init__
    super().__init__(
  File "/usr/local/lib/python3.11/dist-packages/aeon/classification/deep_learning/base.py", line 63, in __init__
    super().__init__()
  File "/usr/local/lib/python3.11/dist-packages/aeon/classification/base.py", line 86, in __init__
    super().__init__()
  File "/usr/local/lib/python3.11/dist-packages/aeon/base/_base_collection.py", line 44, in __init__
    _check_estimator_deps(self)
  File "/usr/local/lib/python3.11/dist-packages/aeon/utils/validation/_dependencies.py", line 351, in _check_estimator_deps
    compatible = compatible and _check_python_version(obj, severity=severity)
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/aeon/utils/validation/_dependencies.py", line 289, in _check_python_version
    raise ModuleNotFoundError(msg)
ModuleNotFoundError: IndividualInceptionClassifier requires python version to be <3.12, but system python version is 3.11.0rc1 (main, Aug 12 2022, 10:02:14) [GCC 11.2.0].

Versions

Python 3.11.0rc1 (main, Aug 12 2022, 10:02:14) [GCC 11.2.0] on linux Type "help", "copyright", "credits" or "license" for more information. >>> from aeon import show_versions;show_versions()

System:
python: 3.11.0rc1 (main, Aug 12 2022, 10:02:14) [GCC 11.2.0]
executable: /usr/bin/python
machine: Linux-6.5.0-26-generic-x86_64-with-glibc2.35

Python dependencies:
pip: 23.3.1
setuptools: 69.0.1
scikit-learn: 1.4.1.post1
aeon: 0.7.1
statsmodels: None
numpy: 1.26.2
scipy: 1.12.0
pandas: 2.0.3
matplotlib: 3.8.4
joblib: 1.3.2
numba: 0.59.1
pmdarima: None
tsfresh: None

[baraline]

Hi @zobelb I do not seem to be able to reproduce your issue on python 3.11. What is your TensorFlow version ? I'm using 2.16.1. Otherwise, try to update to aeon 0.8 to see if you still get the issue.

from aeon.datasets import load_unit_test
from aeon.classification.deep_learning import InceptionTimeClassifier

X_train, y_train = load_unit_test(split="train")
X_test, y_test = load_unit_test(split="test")

inctime = InceptionTimeClassifier(n_epochs=20,batch_size=4)
print(inctime.fit(X_train, y_train))

System config :

System:
    python: 3.11.8 | packaged by conda-forge | (main, Feb 16 2024, 20:40:50) [MSC v.1937 64 bit (AMD64)]
executable: C:\Users\antoine\anaconda3\envs\test_env\python.exe
   machine: Windows-10-10.0.22631-SP0

Python dependencies:
          pip: 24.0
   setuptools: 69.2.0
 scikit-learn: 1.4.2
         aeon: 0.8.0
  statsmodels: None
        numpy: 1.26.4
        scipy: 1.12.0
       pandas: 2.0.3
   matplotlib: None
       joblib: 1.4.0
        numba: 0.59.1
     pmdarima: None
      tsfresh: None

[zobelb]

Antoine, this is with the Tensorflow Docker containers that are packed with Python 3.11.0rc1. My suspicion is that it's the "rc1" part that is causing the issue from what I can see in the code. I've tried creating a development install of the aeon package on my computer but am running into issues with that so I can't verify my theory.

ERROR: Project file: aeon has a 'pyproject.toml' and its build backend is missing the 'build_editable' hook. Since it does not have a 'setup.py' nor a 'setup.cfg', it cannot be installed in editable mode. Consider using a build backend that supports PEP 660.

I thought setuptool 69.2.0 should get rid of that error.

[baraline]

You get this error while trying to install aeon from sources right ? This should indeed not happen, which command are you using ?

I'll take a look at the rc1 thing, maybe @MatthewMiddlehurst knows more or maybe @hadifawaz1999 has encountered similar issues during his work on DL ?

[zobelb]

Following the instructions here: https://www.aeon-toolkit.org/en/stable/developer_guide/dev_installation.html

[MatthewMiddlehurst]

A bit strange that an old release candidate is being used, but I'm sure there's a valid reason. Not sure if the function for version checking can handle that, I agree.

AFAIK setuptools>=64 supports PEP 660? The install failure seems a bit weird. If you raise the build-system bound in pyproject.toml does that change anything?

[zobelb]

I got my setup fixed and verified it is way the version is checked.

Here's the fix I implemented for my project. I'm unsure of the proper way to contribute this (haven't finished reading the dev guide yet!)

aeon/aeon/utils/validation/_dependencies.py: _check_python_version has the following lines:

    if sys_version in est_specifier:
        return True

to check the version.
in in this case is equivalent to SpecifierSet.contains:

def contains(self, item: str | Version) -> bool:
"""Return whether or not the item is contained in this specifier.

    :param item: The item to check for.

    This is used for the ``in`` operator and behaves the same as
    :meth:`contains` with no ``prereleases`` argument passed

In order to consider a release candidate valid, we need prereleases to be passed, so I've changed the line locally to:

if est_specifier.contains(sys_version, prereleases=True):
    return True

[baraline]

Good job !
If you want to contribute this change you are welcome to ! Just tell us wheter you would want to do it (so we can implement the fix ourselves if you don't). If you need some help on contributing don't hesitate to message us on Slack.

[zobelb]

I unfortunately don't have a non-work setup in which to do this in the near future.