for example for very big sparse data for binary target/labels is it possible to speed up logistic regression with regularization?