Summary

This chapter focused on two feature learning tools: RBM and word embedding processes.

Both of these processes utilized deep learning architectures in order to learn new sets of features based on raw data. Both techniques took advantage of shallow networks in order to optimize for training times and used the weights and biases learned during the fitting phase to extract the latent structure of the data. 

Our next chapter will showcase four examples of feature engineering on real data taken from the open internet and how the tools that we have learned in this book will help us create the optimal machine learning pipelines.