Scientific Visualization: Python + Matplotlib

Authors: Nicolas P. Rougier Year: 2021 Tags: scientific-visualization, matplotlib, python, data-visualization, tutorial-book, publication-quality-figures

TL;DR

Open-access textbook (CC BY-NC-SA 4.0) covering the full Matplotlib stack for publication-quality scientific figures, from figure anatomy and coordinate transforms through design principles, 3D/animation, and a curated showcase. Targets intermediate Python users who need fine-grained control over figures rather than quick exploratory plotting.

First pass — the five C's

Category. Instructional textbook / practitioner reference; not a research paper reporting original experiments or theory.

Context. Python scientific visualization ecosystem, 2021; situates itself within the landscape mapped by Jake Vanderplas (adapted diagram included); builds directly on John D. Hunter's Matplotlib (first release 2003, v3.4 current at time of writing); Volume II of a "Scientific Python" series; author also co-created VisPy and glumpy.

Correctness. Load-bearing assumptions: (1) Matplotlib is the appropriate primary tool for publication-quality 2D scientific figures; (2) readers have intermediate Python and beginner NumPy; (3) desktop/print use cases dominate over browser/interactive. All three are reasonable but narrow the book's scope explicitly.

Contributions. - Systematic exposition of Matplotlib's four coordinate systems (FC, NFC, DC, NDC) and how to compose/invert their transforms for precise element placement. - Design-principle chapter ("Ten simple rules") and defaults/styling chapter aimed at improving default figure quality, not just API coverage. - Advanced showcase section with full annotated source code for non-trivial figures (domain coloring, Escher projections, self-organizing maps, etc.). - Open-access release with all source code on GitHub, under CC BY-NC-SA 4.0 / BSD-2-clause.

Clarity. Writing is concise and technically precise; code snippets are directly executable with stated imports; a few sections in the provided excerpt end mid-sentence due to OCR/extraction artifacts, which is an artifact of the excerpt, not the original book.

Second pass — content

Main thrust: Progressively builds Matplotlib expertise from the object hierarchy (Figure → Axes → Axis → Artist) through coordinate transforms, scales, projections, design principles, and advanced topics, with every concept accompanied by minimal reproducible code and a labeled source file.

Supporting evidence: - Figure dimensions and DPI: default dpi=100 yields 600×600 px for a 6×6 in figure; publishers require 300–600 dpi; a 4.25 in wide figure at 600 dpi yields 2550 px width. - Version environment stated explicitly: Python 3.7.4, NumPy 1.16.4, SciPy 1.3.0, Matplotlib 3.1.0 (as of June 2019 at time of writing; book published November 2021). - Coordinate transform verification: log-scale equidistance confirmed numerically (all three intervals = 155.0 figure units); linear scale shows 4.185 / 41.85 / 418.5 (factor-of-10 increments). - Five renderer types (Agg/PS/PDF/SVG/Cairo) and six interactive interface types documented in Tables 1.1–1.2. - LaTeX point = 1/72.27 in; Matplotlib point = 1/72 in — noted explicitly as a discrepancy relevant to font-size matching.

Figures & tables: Figures carry source-file citations (e.g., anatomy/anatomy.py), making them reproducible. Axes are labeled in all example figures. No error bars, confidence intervals, or statistical significance — none are appropriate for a textbook. Tables 1.1 and 1.2 are simple lookup tables, clearly structured. Some figure captions in the OCR'd text are partially garbled (Unicode escape sequences visible in scales/projections chapter).

Follow-up references: - Matplotlib official documentation / transformation tutorial — repeatedly cited as the authoritative deep-dive on backends and transforms. - Jake Vanderplas's Python visualization landscape (2018) — source of the ecosystem map reproduced in the Introduction. - pyviz.org — described as maintaining an up-to-date active-package list beyond what the book can track. - Bibliography listed in TOC (pp. 221+) but not included in the provided excerpt; specific cited works not stated here.

Third pass — critique

Implicit assumptions: - Matplotlib alone is sufficient for the vast majority of scientific publishing needs; interactive/web scenarios are acknowledged but not taught. - Static print/PDF is the primary output target; figures optimized for screen (responsive sizing, tooltips) are out of scope. - Code version pinning (Matplotlib 3.1.0, Python 3.7) may silently break for readers using current releases (3.8+), since the API has changed (e.g., ScaledTranslation import path, subplot_kw dict syntax shown uses = inside a dict literal, which is a Python syntax error — {"aspect"=1} should be {"aspect": 1}).

Missing context or citations: - No comparison of Matplotlib's coordinate/transform system against equivalent mechanisms in Bokeh, Plotly, or Altair, even though these are named in the introduction. - No discussion of accessibility (colorblind-safe palettes, screen-reader compatibility) despite a full chapter on colors. - The "ten simple rules" chapter is referenced but not included in the provided excerpt; its relationship to prior visualization guidelines literature (e.g., Tufte, Cleveland) is not stated in the excerpt.

Possible experimental / analytical issues: - Not a research paper; no experiments, baselines, or datasets to critique. - The Python syntax error in subplot_kw={"aspect"=1} (Chapter 1) suggests at least one code snippet was not mechanically tested before publication. - Version mismatch: code written for Matplotlib 3.1.0 (June 2019) but book released November 2021, when Matplotlib 3.4 was current; no compatibility notes provided. - OCR/extraction of the HAL PDF introduces Unicode escape artifacts in figures (Chapter 3 scale comparison figures are unreadable in the provided text).

Ideas for future work: - Add a compatibility matrix or version-conditional code blocks for Matplotlib ≥ 3.5, where several APIs changed. - Extend the showcase to include interactive figures (ipywidgets + Matplotlib) to bridge the gap with browser-native tools mentioned in the introduction. - Add a chapter on accessibility: colorblind-safe palette selection, hatching as a redundant encoding channel, and alt-text generation for figures. - Provide a CI-tested repository that runs all source files against the current Matplotlib release and flags breaking changes automatically.