Hello @nd26! Thanks for updating the PR.

• In the file mlxtend/feature_selection/sequential_feature_selector.py, following are the PEP8 issues :

Line 21:80: E501 line too long (91 > 79 characters)
Line 24:1: E302 expected 2 blank lines, found 1
Line 28:80: E501 line too long (105 > 79 characters)
Line 30:80: E501 line too long (95 > 79 characters)
Line 30:96: W291 trailing whitespace
Line 33:80: E501 line too long (92 > 79 characters)
Line 111:78: W291 trailing whitespace
Line 112:77: W291 trailing whitespace
Line 113:77: W291 trailing whitespace
Line 138:30: E251 unexpected spaces around keyword / parameter equals
Line 138:32: E251 unexpected spaces around keyword / parameter equals
Line 164:80: E501 line too long (86 > 79 characters)
Line 166:80: E501 line too long (87 > 79 characters)
Line 169:9: E303 too many blank lines (2)

Comment last updated on November 01, 2017 at 15:51 Hours UTC

Thanks for the PR! Based on the test results, it just seems to be a Tab vs. Whitespace error. I.e., try to replace tabs by 4 whitespaces each.

Btw. you can always test your code locally by running nosetests mlxtend/feature_selection/, for example.

No problem! Thanks for the reply. Will do tomorrow morning :)

Sure, no hurry!

Btw. added an issue regarding the random number generator you mentioned via email (#274). I am not sure what's more convenient, adding the Repeated CV feature or the random seed first. But I think adding repeated CV (using the current cross_val_score implementation) should work without an extra random seed. After that is implemented, I am happy to rewrite the code to support random seeds.

However, I just wanted to mention it in case the repeated CV causes troubles without random seeds (i.e., irreproducible unit tests). In that case, we may want to add random seeds first and then the repeated CV.

Anyways, thanks a lot for contributing, both repeated CV and the random seeds are useful enhancements!

Most likely, if there are unit tests that test the n_cv_repeats parameter, they will fail due to the lack of reproducibility (SFS is highly unstable when used against a large number of features). But, current tests only test cross_val_score without using different values for n_cv_repeats, right?

No problem man!

Coverage decreased (-0.08%) to 90.828% when pulling 2a5c9c5 on nd26:feature/RepeatedSKFold271 into 3028be8 on rasbt:master.

I just see the reason why one of the unit tests is failing is because one of them is running sklearn 0.18, which doesn't yet have the RepeatedKFold class (it was added in sklearn 0.19).

I think it's fair to switch all the unit tests to sklearn >= 0.19 as it has been out now for a while.

The way to fix this would be to change SKLEARN_VERSION="0.18" in the .travis.yml file to SKLEARN_VERSION="0.19" in the section

- python: 3.5
          env: LATEST="false" COVERAGE="false" NUMPY_VERSION="1.11.2" SCIPY_VERSION="0.18.1" SKLEARN_VERSION="0.18" PANDAS_VERSION="0.19.1"