Add decision function method for stacking classifiers

[qiagu]

Description

Related issues or pull requests

Pull Request Checklist

• Added a note about the modification or contribution to the ./docs/sources/CHANGELOG.md file (if applicable)
• Added appropriate unit test functions in the ./mlxtend/*/tests directories (if applicable)
• Modify documentation in the corresponding Jupyter Notebook under mlxtend/docs/sources/ (if applicable)
• Ran PYTHONPATH='.' pytest ./mlxtend -sv and make sure that all unit tests pass (for small modifications, it might be sufficient to only run the specific test file, e.g., PYTHONPATH='.' pytest ./mlxtend/classifier/tests/test_stacking_cv_classifier.py -sv)
• Checked for style issues by running flake8 ./mlxtend

[pep8speaks]

Hello @qiagu! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file mlxtend/classifier/tests/test_stacking_classifier.py:

Line 555:1: W293 blank line contains whitespace

• In the file mlxtend/classifier/tests/test_stacking_cv_classifier.py:

Line 599:1: W293 blank line contains whitespace

Comment last updated at 2019-12-07 19:18:05 UTC

[qiagu]

The roc_auc score calculated in scikit-learn v0.20.3 is different from the one in v0.21.3. I couldn't figure out what parameter caused the discrepancy.

[coveralls]

Coverage decreased (-0.005%) to 92.349% when pulling d4b550f on qiagu:stacking into 9cb5f6c on rasbt:master.

[rasbt]

Thanks for the PR. Looks pretty good so far!

Regarding

The roc_auc score calculated in scikit-learn v0.20.3 is different from the one in v0.21.3. I couldn't figure out what parameter caused the discrepancy.

I found this in the changelog (https://scikit-learn.org/dev/whats_new/v0.21.html#sklearn-svm):

Maybe you can

1. add a new test for SVC(... decision_function_shape='ovo')
2. change the current test such that it only runs in sklearn > 0.21.3

E.g.,

from distutils.version import LooseVersion as Version
from sklearn import __version__ as sklearn_version
...

def test ...():
    # test code
    if Version(sklearn_version) > Version("0.21.3"):
        assert ...
    else:
        assert ...

[qiagu]

@rasbt Thanks. Your suggestion always makes the most sense. I stop trying other things.

[qiagu]

@rasbt Do we need to care about the minor decrease of test coverage?

[rasbt]

@rasbt Do we need to care about the minor decrease of test coverage?

No worries, that's fine. The change is so small that it is basically non-existent. I am currently a bit swamped with deadlines but let me go over the code again carefully on the upcoming weekend. Thanks for the PR so far!

[rasbt]

Looks great! I just added some examples to the jupyter notebook. Once the test passed, it should be good to merge. Many thanks!

[qiagu]

Thanks for merging. @rasbt