In this chapter, we tapped into a very powerful but lesser known paradigm in machine learning—semi-supervised learning. We began by exploring the underlying concepts of transductive learning and self-training, and improved our understanding of the latter class of techniques by working with a naïve self-training implementation.
We quickly began to see weaknesses in self-training and looked for an effective solution, which we found in the form of CPLE. CPLE is a very elegant and highly applicable framework for semi-supervised learning that makes no assumptions beyond those of the classifier that it uses as a base model. In return, we found CPLE to consistently offer performance in excess of naïve semi-supervised and supervised implementations, at minimal risk. We've gained a significant amount of understanding regarding one of the most useful recent developments in machine learning.
In the next chapter, we'll begin discussing data preparation skills that significantly increase the effectiveness...