recommenders-team / recommenders

Best Practices on Recommendation Systems
https://recommenders-team.github.io/recommenders/intro.html
MIT License
18.71k stars 3.06k forks source link

[BUG] inconsistent dependency on Pymanopt #2038

Open SimonYansenZhao opened 9 months ago

SimonYansenZhao commented 9 months ago

Description

In setup.py, pymanopt is required as a dependent package:

extras_require["experimental"] = [
    # xlearn requires cmake to be pre-installed
    "xlearn==0.40a1",
    # VW C++ binary needs to be installed manually for some code to work
    "vowpalwabbit>=8.9.0,<9",
    # nni needs to be upgraded
    "nni==1.5",
    "pymanopt>=0.2.5",
]

However, in tests/ci/azureml_tests/submit_groupwise_azureml_pytest.py, pymanopt is replaced:

    conda_dep.add_pip_package(
        "pymanopt@https://github.com/pymanopt/pymanopt/archive/fb36a272cdeecb21992cfd9271eb82baafeb316d.zip"
    )

This inconsistency may cause problems in the future, and complicate testing.

In which platform does it happen?

This behavior only happens in testing.

How do we replicate the issue?

Triggering a test will make it happen.

Expected behavior (i.e. solution)

Testing should not change the dependencies of the recommenders package.

Other Comments

SimonYansenZhao commented 9 months ago

Some modules were renamed in the new version of Pymanopt, such as pymanopt.solvers. See https://github.com/pymanopt/pymanopt/issues/207

miguelgfierro commented 8 months ago

Related to https://github.com/recommenders-team/recommenders/issues/1943