Fix VIF add a new function 'feature_selection_vif'

[pydemia]

1. Fix 'variance_inflation_factor' in stats.outliers_influence : Apply sm.add_constant process for proper results.
2. Add 'feature_selection_vif' in stats.multivariate_tools : A Function of Stepwise Feature Elimination with VIF

[coveralls]

Coverage increased (+0.0002%) to 88.026% when pulling be05459 on pydemia:pydemia into 22f2766 on statsmodels:master.

[pydemia]

The Travis CI throws 2 errors on Python 2.6, which are irrelevant to my commit :

1. KeyError: 'Canonical Correlation'
   ERROR: statsmodels.multivariate.tests.test_cancorr.test_cancorr

2. KeyError: 'Wilks\xe2\x80\x99 lambda'
   ERROR: statsmodels.multivariate.tests.test_manova.test_manova_sas_example
   ERROR: statsmodels.multivariate.tests.test_manova.test_manova_test_input_validation
   ERROR: statsmodels.multivariate.tests.test_multivariate_ols.test_glm_dogs_example
   ERROR: statsmodels.multivariate.tests.test_multivariate_ols.test_specify_L_M_by_string
   ERROR: statsmodels.multivariate.tests.test_multivariate_ols.test_from_formula_vs_no_formula

[coveralls]

Coverage increased (+0.0002%) to 88.026% when pulling 176d5bf on pydemia:pydemia into 22f2766 on statsmodels:master.

[coveralls]

Coverage decreased (-5.2%) to 82.853% when pulling bba4762 on pydemia:pydemia into 22f2766 on statsmodels:master.

[josef-pkt]

Sorry for the slow response. I need to think about this and it's too far away from what I'm currently looking at.

e.g. adding the sm.add_constant is not backwards compatible and will most likely break current use cases.

[josef-pkt]

The feature_selection_vif looks fine in general. There should be no print in the code.

Unit tests are missing.

[josef-pkt]

This pull request introduces 1 alert when merging 176d5bf into 22f2766 - view on lgtm.com

new alerts:

• 1 for Variable defined multiple times

---

Comment posted by lgtm.com

[pydemia]

I should apply your mentions and retry the pull request.