Machine Learning Algorithms

 Data conversion

This step is probably the simplest and, at the same time, the most important when handling categorical data. We have discussed several methods to encode labels using numerical vectors and it's not necessary to repeat the concepts already explained. A general rule concerns the usage of integer or binary values (one-hot encoding). The latter is probably the best choice when the output of the classifier is the value itself, because, as discussed in Chapter 3, Feature Selection and Feature Engineering, it's much more robust to noise and prediction errors. On the other hand, one-hot encoding is quite memory-consuming. Therefore, whenever it's necessary to work with probability distributions (like in NLP), an integer label (representing ...