Update to work with Python 3.8 and sklearn >0.20.0

[john-sandall]

I came across some issues/warnings when using Python 3.8 and more recent versions of sklearn when debugging solutions to #11

For example, using Python 3.8 with scikit-learn 0.20.0 you get a couple issues:

1. Warning - sklearn 0.20.0 uses an version of joblib/cloudpickle that uses imp instead of importlib

/Users/john/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/externals/cloudpickle/cloudpickle.py:47: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp

imp has been deprecated since Python 3.4 but I think the warning maybe only shows on Python 3.8? Relating the sklearn, this is well documented in scikit-learn/scikit-learn#12226 and scikit-learn/scikit-learn#12434 and solved I think in 0.20.1 onwards.

2. TypeError: an integer is required (got type bytes)

This is another incompatibility between Python 3.8 and the bundled versions of joblib/cloudpickle in sklearn 0.20.0, similar to this.

Full traceback:

In [6]: import sklearn
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-6-b7c74cbf5af0> in <module>
----> 1 import sklearn

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/__init__.py in <module>
     63     from . import __check_build
     64     from .base import clone
---> 65     from .utils._show_versions import show_versions
     66
     67     __check_build  # avoid flakes unused variable error

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/utils/__init__.py in <module>
     11
     12 from .murmurhash import murmurhash3_32
---> 13 from .validation import (as_float_array,
     14                          assert_all_finite,
     15                          check_random_state, column_or_1d, check_array,

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/utils/validation.py in <module>
     25 from ..exceptions import NotFittedError
     26 from ..exceptions import DataConversionWarning
---> 27 from ..utils._joblib import Memory
     28 from ..utils._joblib import __version__ as joblib_version
     29

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/utils/_joblib.py in <module>
     16         from joblib import logger
     17 else:
---> 18     from ..externals.joblib import __all__   # noqa
     19     from ..externals.joblib import *  # noqa
     20     from ..externals.joblib import __version__  # noqa

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/__init__.py in <module>
    117 from .numpy_pickle import load
    118 from .compressor import register_compressor
--> 119 from .parallel import Parallel
    120 from .parallel import delayed
    121 from .parallel import cpu_count

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/parallel.py in <module>
     30                                  LokyBackend)
     31 from ._compat import _basestring
---> 32 from .externals.cloudpickle import dumps, loads
     33 from .externals import loky
     34

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/externals/cloudpickle/__init__.py in <module>
      1 from __future__ import absolute_import
      2
----> 3 from .cloudpickle import *
      4
      5 __version__ = '0.5.6'

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/externals/cloudpickle/cloudpickle.py in <module>
    149
    150
--> 151 _cell_set_template_code = _make_cell_set_template_code()
    152
    153

~/.virtualenvs/sklearn-genetic/lib/python3.8/site-packages/sklearn/externals/joblib/externals/cloudpickle/cloudpickle.py in _make_cell_set_template_code()
    130         )
    131     else:
--> 132         return types.CodeType(
    133             co.co_argcount,
    134             co.co_kwonlyargcount,

TypeError: an integer is required (got type bytes)

[john-sandall]

Digging into things, scikit-learn 0.22 is the first version to explicitly work with Python 3.8: https://github.com/scikit-learn/scikit-learn/blob/0.22/setup.py

This is also confirmed here: scikit-learn/scikit-learn#15328 (comment)

I'm submitting a PR which will use environment markers (see PEP 508 and usage example for setup.py as follows to control for requiring sklearn>=0.22 for Python 3.7 only:

install_requires=[
        "scikit-learn>=0.20;python_version<'3.8'",
        "scikit-learn>=0.22;python_version>='3.8'",
        "deap>=1.0.2",
    ],

[john-sandall]

Another fun issue is that SelectorMixin exists in different places:

• In <0.22 it's in feature_selection.base
• In 0.22 it's in feature_selection._base
• In 0.23 it's in the public API: https://scikit-learn.org/dev/whats_new/v0.23.html#sklearn-feature-selection

PR on its way that fixes all of this.

[manuel-calzolari]

Closed with commit c029c0e. I didn't add the scikit-learn>=0.22 requirement for Python >= 3.8 because I think this is a scikit-learn issue.