Description

• The solvers are {newton-cg, lbfgs, liblinear, sag}, default: liblinear
• Only liblinear supports L1 penalty
• multi_class='multinomial' "Works only for the newton-cg, sag and lbfgs solvers."

http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html

I'm not an expert... Would it be possible to add L1 to any of these other solvers or is that mathematically impossible?