fix(multiprocessing): Use list instead of key-view

[eddiebergman]

Seems some tests were failing stochastically in AutoML::fit() -> _fit_cleanup(), many complaining that self._logger was None and calling methods like info and exception on this None object were failing (as expected). Scrolling further up in the stack traces, there was a seemingly unrelated failure.

     all_loaded_modules = sys.modules.keys()
     preload = [
        loaded_module
        for loaded_module in all_loaded_modules
        if loaded_module.split(".")[0]
        in (
            "smac",
            "autosklearn",
            "numpy",
            "scipy",
            "pandas",
            "pynisher",
            "sklearn",
            "ConfigSpace",
        )
        and "logging" not in loaded_module
    ]
    # RuntimeError: dictionary changed size during iteration

My guess here is that sys.modules was slowly getting cleaned up by the garbage collector as cleanup was happening, causing this iteration to fail. Solution was just to wrap in a list as list(sys.modules.keys()).

Will update depending on whether this fixes the issue.

[codecov]

Codecov Report

Merging #1609 (045fb68) into development (6a97f72) will decrease coverage by 0.09%.
The diff coverage is 100.00%.

Additional details and impacted files

@@               Coverage Diff               @@
##           development    #1609      +/-   ##
===============================================
- Coverage        83.32%   83.23%   -0.10%     
===============================================
  Files              156      156              
  Lines            11927    11927              
  Branches          1896     1896              
===============================================
- Hits              9938     9927      -11     
- Misses            1422     1431       +9     
- Partials           567      569       +2