Hands-On Unsupervised Learning with Python

Semi-supervised learning algorithms

A semi-supervised scenario can be considered as a standard supervised one that exploits some features belonging to unsupervised learning techniques. A very common problem, in fact, arises when it's easy to obtain large unlabeled datasets but the cost of labeling is very high. Hence, it's reasonable to label only a fraction of the samples and to propagate the labels to all unlabeled ones whose distance from a labeled sample is below a predefined threshold. If the dataset has been drawn from a single data generating process and the labeled samples are uniformly distributed, a semi-supervised algorithm can achieve an accuracy comparable with a supervised one. In this book, we are not discussing these algorithms; ...