I was surprised to see a box at my front door. I had not ordered anything… or had I? Well, it turns out it was a box I was indeed expecting, but not until later in the month. I was then doubly surprised when I opened the box and found a familiar spine, cover and a new barcode. The physical copies of the second edition of Advanced Data Science and Analytics with Python arrived at my front door yesterday, and I have to report that holding a book you wrote is still stranger than the first time.
The second edition is not a reprint with a new cover and a tidier index. The landscape shifted underneath the first edition in ways that would have sounded like speculative fiction when it went to press. Generating code is now close to free (still watch your token budget!). Looking at a model output critically, deciding whether it is ready for a production pipeline or yet another interesting experiment: none of that got cheaper. If anything it got scarcer, precisely because the cheap part became so abundant that it now hides the expensive part. It is the same barrier I keep running into in the boardroom, incidentally, and it is rarely the model.
So the book leans into that gap. There is expanded treatment of neural networks and deep learning, a heavier hand on natural language processing, more room for graph and social network analysis, a proper look at getting machine learning models into production rather than into a notebook, and a new chapter on generative AI that tries to be useful in eighteen months’ time rather than merely current today. Which, as anyone writing about this field knows, is a bet placed with a bookmaker who does not publish odds.
What has not changed is the tone. This was never intended to be a cookbook, and the second edition resists that urge as firmly as the first did. The recipes are available everywhere. What is harder to find is the reasoning that tells you which recipe belongs in front of you, what it assumes about your data, and how it fails when those assumptions quietly stop holding. Python remains the vehicle throughout, not the destination.
If you have been following the argument here about judgement being the scarce input in data science, the book is the long-form version of it. I wrote more about that framing in a companion post on the well-posedness of prompts, and the two are best read as a pair.
Copies are making their way out into the world, with the official publication date on 26 August. Details of this and the companion volume are on the books page. If one reaches your desk, I would be glad to hear which chapter provoked the most disagreement. Disagreement is the useful signal. Agreement mostly means I wrote something you already believed.
