Machine Learning Algorithms - Second Edition

Normalization and regularization

Normalizing a numeric dataset is one of the most important steps, particularly when different features have different scales. In Chapter 3, Feature Selection and Feature Engineering, we discussed several methods that can be employed to solve this problem. Very often, it's enough to use StandardScaler to whiten the data, but sometimes it's better to consider the impact of noisy features on the global trend and use RobustScaler to filter them out without the risk of conditioning the remaining features. Moreover, there are many algorithms that can benefit from whitened datasets; therefore, I suggest using a grid search (we're going to discuss how to merge different transformations later in this chapter) and pick ...