Standarization (Feature Scaling)

[jsjason]

For lasso regression (and potentially other algorithms), we normally perform feature scaling on the input data before feeding it to the algorithm. However, feature scaling is not trivial in distributed environments (data parallelism across many workers) and may, in fact, hinder the convergence of the algorithm if applied incorrectly. Thus, we might want to exclude feature scaling for now, especially if its relevance is not so great for the algorithm.