I want to run a regularized logistic regression on a dataset with 25 million observations and about a 1000 mostly non-sparse columns with non-ignorable weights.

My first choice would be BayesGLM, but I'm sure that isn't remotely close to computationally feasible. I was looking into the LASSO/ElasticNet implementations in SciPi and it doesn't seem to have the ability to take weighted data. Does anybody have any recommendations?

  • $\begingroup$ glmnet $\endgroup$ – Cyan Sep 20 '13 at 0:18
  • $\begingroup$ I'm not really sure that R can really handle data that size particularly well $\endgroup$ – DavidShor Sep 23 '13 at 10:33
  • $\begingroup$ Hmm... biglars uses the transparent-flat-file-storage package ff to get around that difficulty. Maybe we need a bigglmnet package. I don't know your R hacking skill level; it's within my capabilities, but I have other things on my to-do list right now... $\endgroup$ – Cyan Sep 23 '13 at 16:30
  • $\begingroup$ The latest Liblinear can handle this problem easily. you can search liblinear via Google. $\endgroup$ – user30839 Sep 29 '13 at 9:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.