GLMNET is an R package which implements a fast algorithm for estimation of generalized linear models with convex penalties, such as a linear regression problem. The package isn't able to handle discrete choice multinomial regression.
I want an implementation which is able to handle discrete choice multinomial logistic regression, must be faster than a python numpy implementation and also be straightward to interface/implement with/in python.
Here are important links which will clarify the problem.
Model: [url removed, login to view]
GLMNET paper: [url removed, login to view]~hastie/Papers/[url removed, login to view]
Optimization routine: [url removed, login to view]