Random subsampling for feature selection #47

[xliu833]

Description

Related issues or pull requests

Fixes #47

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

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2019-12-21 22:28:28 UTC

[rasbt]

Hi Lareina, thanks for the PR. I don't quite understand how the value is chosen k_to_select = int(len(X[1])**.5). So this means that the random subset is always sqrt of the original number of features?

There are two problems with that.

1. We want to add the random feature subset size as an optional thing
2. I think we should let the user allow to choose the subset size

There could be a new parameter

use_random_feature_subset for the SequentialFeatureSelector class which either accepts a function like f = lambda x: int(np.sqrt(x) or None.

Let me know if you have questions.