I opened my mailbox late this morning and there it was, an email from CRC Press / Taylor & Francis by editor’s office, the gist: the book has been sent to the printers. Hurrah!
Publication date: 26 August 2026. Pre-orders open 5 August 2026. Copies ship after the 26th. You can get it here.
If you have been following along, you will know this has been a slow-burn serial rather than a surprise release. Seventeen months ago I announced I had started work on the second edition. Six weeks after that, I reported back that there was, in fact, rather a lot to do — a sentence I wrote with more optimism than the evidence warranted. In January the cover arrived and as promised in blog post the Jackalope returned. In March it entered production. In May I finished the proofs, which is the stage at which a manuscript stops being an idea and becomes an actual object with page numbers. The email I received today means that it is now time to become ink on paper.
Why a second edition, and why now
The first edition was published in May 2020, with the physical copies turning up on my desk a week later and catching me entirely unprepared. Its table of contents was a self-contained, idiosyncratic answer to the question what does “advanced” actually mean? — time series analysis, natural language processing, topic modelling, social network analysis, neural networks and deep learning, and, at the end, deployment of a trained model into an iPhone app so that the thing reached an actual human being’s thumb.
Read that list again in 2026. Most of it is now the middle of the curriculum, not the far edge of it. Topic modelling with LDA remains a perfectly good technique that a great many people have quietly replaced with an embedding and a clustering step. Word vectors have been eaten by attention. “Deploy a model” has acquired a supply chain.
So, as I set out when the proofs were done, the second edition expands substantially into the territory that has genuinely moved: transformers, generative AI and large language models, vector search, graph embeddings, and modern deployment workflows. Not as a tour of the news cycle, but as working code with the mathematics left in.
This is the part where a second edition earns its keep or doesn’t. It is very easy to write a revision that is really a press release: bolt on a chapter called “Generative AI”, wire up an API call, declare the field addressed. That book would be obsolete by Christmas. It would also be teaching the single most expensive habit in contemporary data science, which is the belief that because the abstraction layer got friendlier, the fundamentals stopped applying.
They did not. Attention is still linear algebra with a normalisation trick. A vector database is still an approximate nearest-neighbour index with a marketing budget, and the recall/latency trade-off you are making is precisely the one you would have made with a k-d tree in 2010 — only now it is hidden behind a client library and a monthly invoice. Graph embeddings will happily encode the structural bias in your data and hand it back to you with the confidence of a Turing Award winner. If you cannot say what your model is doing when it is wrong, you do not have a model. You have a slot machine with good manners.
That, more than any individual chapter, is the argument the second edition is making.
With that in mind, in January I compared writing a second edition to rebooting a long-running sci-fi series: preserve the canon, improve the special effects, close the plot holes. I stand by that, with one clarification for the purists: I have not gone back and made Han shoot second.
The structure of the first edition survives. Chapters remain as independent as I could keep them, so the book still works as a reference you open at the point of need rather than a novel you must read from page one. Worked examples still walk from problem to code to result, and the code still runs.
For anyone tracking the pattern, this is the third time in recent years that I have gone back to a first book. And the most recent revision comes witha decade of hindsight as I am reviewing my book Essential MATLAB and Octave in June. Revisiting your own prose at that distance is an odd sort of code review. The past you was confident about several things. And, boy, am I glad I left myself some comments in my own book!
But back to this 2nd edition: The audience for the book is the same: people who can already program, who have met a statistical model, and who now need to do something consequential with both. Practitioners who have to defend a decision to a board. Postgraduates who have discovered that the tutorial and the production system share only a name. Anyone handed a legacy pipeline, a licence renewal and a deadline, and expected to produce insight from all three.
If you have read my thoughts on why AI adoption fails for organisational rather than technical reasons, or on what an obedient model escaping its sandbox actually tells us, you will recognise the temperament: enthusiasm, with the receipts checked.
On pre-ordering
Pre-orders matter more than they look. They are the signal publishers use to set print runs, and they are the difference between a book that is stocked and a book that is available to order. If you were going to buy it in October, buying it in August is the same money doing considerably more work. Last time round my publisher let me offer a discount ahead of publication; I will ask again, and post here if I have anything to share.
Details for all my books live on the Books page, and I will add the pre-order link here on the 5th.
Thank you to Poppy and Randi and the team at CRC Press / Taylor & Francis, to the reviewers who told me the uncomfortable things while there was still time to act on them, and to everyone who read the first edition and then wrote to tell me what was missing. A surprising amount of the second edition is your fault. And as usual, any omissions or errors are mine.
Somewhere there is now a machine putting ink on paper with my name on it. Forty-two days, give or take, until it lands on someone’s desk. Do let me know what you make of it.