Best NumPy Books

A free, full-length book that focuses specifically on the transition from writing Python loops to writing vectorised NumPy code. Rougier's case studies — rebuilding common algorithms as vectorised expressions — are the most effective exercises for developing the NumPy mindset. Nothing else teaches this shift as deliberately.

BeginnerPaid

Python for Data Analysis

Wes McKinney · O'Reilly Media · 471 pages · ~12 hrs to read

McKinney's book opens with two strong chapters on NumPy before moving into pandas. For engineers who will use NumPy primarily as a substrate for data work, this framing — arrays as the building blocks of DataFrames — is exactly the right mental model to establish early.

A free, full-length book that focuses specifically on the transition from writing Python loops to writing vectorised NumPy code. Rougier's case studies — rebuilding common algorithms as vectorised expressions — are the most effective exercises for developing the NumPy mindset. Nothing else teaches this shift as deliberately.

AdvancedPaid

Guide to NumPy

Travis E. Oliphant · Self-published · 364 pages · ~9 hrs to read

Written by the creator of NumPy, this is the closest thing to a specification document the library has. Dense and unapologetically technical — it explains array internals and memory layout decisions that no secondary source gets right, because the author is the one who made them.

IntermediatePaid

Elegant SciPy

Juan Nunez-Iglesias, Stéfan van der Walt & Harriet Dashnow · O'Reilly · 277 pages · ~7 hrs to read

Uses real scientific datasets to teach NumPy-based vectorised thinking through the lens of writing genuinely elegant, idiomatic code rather than just correct code. The gene-expression and image-processing case studies are a step up in difficulty from typical tutorials and reward engineers who already know the basics.

AdvancedPaid

High Performance Python

Micha Gorelick & Ian Ozsvald · O'Reilly0

Not a NumPy book on its face, but its chapters on memory layout, vectorisation, and when NumPy actually beats pure Python are the most rigorous performance treatment available. Engineers who need to explain why a NumPy operation is slow — not just make it faster by guessing — should read this.