Plot Styling

Plot Styling

Choose the right Python plotting library for the chart type, then apply journal-quality defaults so the panel that comes out of savefig is ready for the figures:scientific-figure composer — no manual cleanup, no font-size firefighting, no chart junk.

Two questions, two minutes

1. What chart type am I drawing? That answer picks the library (references/library-decision-tree.md).
2. Which journal am I targeting? That answer picks the style sheet and font minimum (references/sciplots-recipes.md).

Run those choices through the export conventions below and the output SVG should pass figure-qa's plot-script branch and feed scientific-figure without scale-down surprises.

Library decision tree (summary)

Full decision tree with concrete code snippets per branch: references/library-decision-tree.md.

Journal-quality defaults (matplotlib / seaborn)

The cleanest path is to install SciencePlots and apply its style sheet at the top of every plot script. The relevant style names:

• science — base style (sans-serif, no chart junk, tight margins)
• nature — Nature column dimensions and font sizing
• ieee — IEEE narrow-column with grayscale-safe palette
• vibrant / bright / high-contrast — colorblind-safe palettes from Paul Tol
• notebook — slightly larger fonts for screen reading (avoid for paper)

pythonCopy code
import matplotlib.pyplot as plt
import scienceplots  # noqa: F401  (registers the styles)

plt.style.use(["science", "nature", "no-latex"])

no-latex is non-optional unless you have system LaTeX installed AND want path-rendered text (in which case validate_fonts.py cannot inspect font sizes — see the "no-latex" note in references/sciplots-recipes.md).

For a Nature panel at 1-column width with three lines:

pythonCopy code
import numpy as np
import matplotlib.pyplot as plt
import scienceplots  # noqa: F401

plt.style.use(["science", "nature", "no-latex"])

fig, ax = plt.subplots(figsize=(3.5, 2.5))  # 89 mm x ~63 mm
t = np.linspace(0, 1, 200)
for k, label in enumerate(("control", "drug A", "drug B")):
    ax.plot(t, np.sin(2 * np.pi * (k + 1) * t), label=label, linewidth=1.0)
ax.set_xlabel("time (s)")
ax.set_ylabel("amplitude (a.u.)")
ax.legend(frameon=False, loc="upper right")
fig.savefig("panel_a.svg", bbox_inches="tight", transparent=True)

That single style declaration sets sans-serif fonts at journal-appropriate sizes (Nature 5–7 pt range), removes the right and top spines, applies tight margin defaults, and picks a colorblind-safe palette. See references/sciplots-recipes.md for IEEE, Science, and APS variants.

Export conventions

Every plot panel that will be composed by figures:scientific-figure should:

1. Save as SVG. savefig("panel.svg", ...). SVG preserves text as <text> so validate_fonts.py can inspect every label.
2. Use transparent=True. The composer expects panels with transparent backgrounds so they composite cleanly onto the figure canvas. figure-qa's plot-script branch flags transparent=False and missing-transparent as issues.
3. Use bbox_inches='tight'. Trim whitespace at save time so the panel's bounding box matches its visual extent. The composer's add_panel(..., scale=...) math assumes this.
4. Embed text, not paths. Default matplotlib SVG output embeds text. If you ever set svg.fonttype = 'path', font validation breaks because there are no <text> elements to inspect.
5. Size the figure in inches to match the final mm panel size. A Nature 1-column panel is 89 mm = ~3.5 in. figsize=(3.5, ...) produces a panel that the composer can place at scale 1.0 with no font-size shrinkage.

The figure-qa plot-script branch (check_plot_script.py) checks all of these statically from the script's AST. It reports savefig_not_transparent, savefig_missing_bbox_inches, and any rcparam_font_sizes below the journal minimum. See examples/sciplots_panel.py for a script that passes the QA branch cleanly.

When matplotlib defaults bite you

Common matplotlib output failures and the SciencePlots-style fix:

Quality assurance

After authoring a plot script, invoke figures:figure-qa through the host's skill-invocation mechanism (its programmatic branch is the command below):

bashCopy code
SCRIPTS_DIR="${CLAUDE_PLUGIN_ROOT}/agents/figure-qa-scripts"
[ -d "$SCRIPTS_DIR" ] || SCRIPTS_DIR="$(find . -type d -name figure-qa-scripts -path '*/figures/agents/*' | head -1)"
uv run python "$SCRIPTS_DIR/check_plot_script.py" panel.py --journal nature

The plot-script branch detects the libraries used, reports rcParams font sizes (numeric and dynamic), inspects every savefig call, and offers a library-switch recommendation when the chart type would benefit. After running the script for real and producing the SVG, the raster/SVG branches verify the output too.

Additional resources

• references/library-decision-tree.md — full decision tree with concrete snippets per branch (statistical, grammar-of-graphics, interactive, 3D, R-bridge)
• references/sciplots-recipes.md — Nature, IEEE, Science, APS style recipes; per-journal font and color setup
• references/element-plots.md — original matplotlib / seaborn / plotly / ggplot2 element guide ported from the legacy plugin
• examples/matplotlib-element.py — canonical matplotlib panel
• examples/sciplots_panel.py — SciencePlots-styled panel passing figure-qa's plot-script branch