◢◤ ARKANE_
SYS_TIME UTC+0
Work with me

Preface

Solutions Architects sit at one of the most interesting intersections in modern technology: between what customers need and what’s technically possible. For years, that meant knowing cloud infrastructure, distributed systems, and integration patterns. That’s no longer enough. Customers now build machine learning pipelines and ask hard questions that can’t be answered from a whitepaper.

Classical Data Science is Volume 1 of a series written for that reader: someone who already thinks in systems and already knows Python, but needs a rigorous, end-to-end path through data science before tackling modern AI in Volume 2. This book covers the foundations, data work, classical machine learning, evaluation, and project lifecycle that most practising data scientists spend most of their time in. It’s less glamorous than large language models, but it’s where most real problems still get solved.

The AWS Solutions Architect angle is not incidental. Every chapter that has a natural cloud-native counterpart includes an AWS SageMaker Notes sidebar, so the path from “I understand this technique” to “I can deploy this technique on AWS” stays short throughout.

How to Read This Book

Each chapter follows a consistent anatomy, adapted from the Diataxis documentation framework (Procida 2020):

Section Purpose
Motivation The problem that existed before this technique, in plain language
Exposition The mechanism — why the technique works, not just how to invoke it
Trade-offs and Limits Where the technique holds up, and where it quietly fails
Procedures Formally labelled, task-oriented recipes with code listings
Worked Example A narrated run showing the technique applied to a realistic scenario
AWS SageMaker Notes The managed-service equivalent, and when it’s worth the overhead
History and Further Reading Context, and where to go deeper
How This Connects Explicit links to chapters this one builds on and enables
Exercises Difficulty-tagged practice, answers in Appendix B
Mental Model A one-line distillation to carry forward

You don’t have to read a chapter start to finish in one sitting. Many readers read Motivation through Trade-offs first, work through the Worked Example at a computer later, and keep Appendix A open as a desk reference while coding. Quizzes: andrewrkane.com/quiz/vol1.

Notation and Conventions

Formulas are typeset in standard mathematical notation (e.g. y^=σ(w⊤x+b)\hat{y} = \sigma(w^\top x + b)). Code listings are numbered per chapter (Listing 4.3 is the third listing in Chapter 4), and each chapter has a matching notebook in the companion repository, github.com/arkane-dev/book-companion-notebooks.

Exercises are tagged by kind, not just difficulty:

  • [W] Warm-up: checks understanding of a definition or fact; a few minutes.
  • [A] Applied: apply the chapter’s method to new data or parameters; requires running code.
  • [E] Extension: combines this chapter with prior material, or probes a trade-off boundary; open-ended.
  • [R] Research: requires reading outside the chapter; discussion only in Appendix B, no single answer.

Prerequisites Map

This volume assumes comfort with Python (functions, loops, classes, pip) and the command line, but no prior statistics, calculus, or machine learning background. Chapters 1-5 build those foundations. Chapters are meant to be read in order on a first pass; later chapters assume earlier ones.

Procida, Daniele. 2020. Diataxis: A Systematic Framework for Technical Documentation.
online /books/classical-data-science/preface Newcastle, UK v0.1.0 // rss