Scikit-learn: Add support for dropping collinear variables
Describe the workflow you want to enable
Can we add a feature in LinearRegression that could remove collinearity (exact collinearity) in the data?.
Describe your proposed solution
My proposal is to add an extra argument like remove_collinearity if it is set by the user then we can remove exact collinear variables using the rank of the matrix or collinear variables using VIF. This can save some time instead of going for Ridge regression.
divyaprabha123
All 3 comments
It might be better to have this as a prepreprocessor in sklearn.feature_selection, that way it could be applied to multiple estimators. I'm not sure that exact collinearity is a frequent issue though. Maybe an estimator with a user defined feature correlation threshold?
I'm not sure if it's something that is often done, as opposed to say feature clustering? The latter can be done in scikit-learn with cluster.FeatureAgglomeration though maybe the interface with a required n_clusters is not ideal.
cc @glemaitre
rth
on 3 Feb 2020
This is being worked on as a feature selection transformer here: https://github.com/scikit-learn/scikit-learn/pull/14698
thomasjpfan
on 3 Feb 2020
Indeed thanks. Closing this issue as a duplicate of https://github.com/scikit-learn/scikit-learn/issues/13405 then. If you have other comments or suggestions @divyaprabha123 please comment there.
rth
on 3 Feb 2020
Related issues
amueller
·
3Comments
rebeccaroisin
·
4Comments
genliu777
·
4Comments
yinruiqing
·
3Comments
stephantul
·
3Comments
Most helpful comment
This is being worked on as a feature selection transformer here: https://github.com/scikit-learn/scikit-learn/pull/14698