Engineer picks 3 English MLOps books, led by Machine Learning Engineering with Python

The author's own story sets the frame: a model that looked accurate in a notebook began producing clearly wrong predictions within a week of release, and the cause turned out to be a mismatch between how features were built during training and during inference. That is an engineering problem rather than a modeling one, and it is the gap the three books are meant to close. The stated method was to switch from thick standard volumes — repeatedly bought and left unread — to books of around 300 pages that could be finished and read in sequence in a short period.

Each book plays a different role. Andrew P. McMahon's Machine Learning Engineering with Python, written from the author's experience leading machine learning engineers in the UK, walks through requirements, training pipelines, packaging, deployment and monitoring using tools that appear in real work such as pytest, MLflow, Airflow and containers, and it also covers team structure and how projects are run. The Kelleher brothers' Machine Learning in Production starts from software engineering basics like version control, testing and algorithmic complexity, then moves through probability, statistics, regression and dimensionality reduction into designing data products. Matt Harrison's Machine Learning Pocket Reference narrows to structured data — missing values, feature creation, imbalanced data, model choice, evaluation metrics and feature importance — and is presented as something to open during work rather than read cover to cover.

The author's suggested progression is the three in order, which he says minimizes overlap, and he notes the combined length is about that of a single thick standard book. For anyone who has accumulated unread volumes, the argument is that splitting the material into finishable units makes it far less likely the reading stalls.