Scikit-Learn’s Defaults are Wrong

A very interested read that I missed previously. Started with a tweet from the author that erupted into a discussion about how logistic regression in Scikit-Learn uses L2 penalization as default options.

Far from bashing Scikit-Learn, the article explores interesting ideas: the rational behind default arguments for an API and why the default choice of regularization can be dangerous in the case of logistic regression.

But even if we accept for the sake of argument that most or all of your logistic regressions should be penalized, is there reason to believe that this default functionality is bad? Yes, absolutely.


Want to receive more content like this in your inbox?