Hacky solution: On a suitably equipped unix-like system one might...

  mkdir dlbook;cd dlbook;wget --recursive --level=1 http://www.deeplearningbook.org/

  cd www.deeplearningbook.org/contents

  python
  import pdfkit
  pdfkit.from_file(
  ["TOC.html","acknowledgements.html","notation.html","intro.html","part_basics.html","linear_algebra.html","prob.html","numerical.html","ml.html","part_practical.html","mlp.html","regularization.html","optimization.html","convnets.html","rnn.html","guidelines.html","applications.html","part_research.html","linear_factors.html","autoencoders.html","representation.html","graphical_models.html","monte_carlo.html","partition.html","inference.html","generative_models.html","bib.html","index-.html"], 
  "Goodfellow-et-al-2016-Book.pdf")

Better solutions?

edit: changed ".pdf" from a slightly longer approach to ".html" which actually exists in this workflow. Thanks @TheCabin!

edit(2): ... and gave a valid path to this listdir. Check before I post... check before I post...

edit(3): ... and removed the os.listdir line no longer needed in this approach. Gosh. Just ignore what I was saying and build your own approach. That'll probably be faster at this rate.

edit(4): Don't import os without using it.