data-visualization_skill

Overview

This skill provides practical best practices for creating clear, accurate scientific visualizations in Python using matplotlib, seaborn, and related libraries. It focuses on avoiding common pitfalls (especially with log scales), optimizing layout and element sizes for publication, and preparing images that meet Claude API size constraints. The guidance is concise, actionable, and oriented to reproducible, publication-quality figures.

How this skill works

The skill inspects common visualization choices and points out technical issues that can mislead readers (for example, KDE-based violin plots on log axes). It recommends concrete alternatives and code patterns: transform data before plotting, use boxplots or histograms on log scales, size elements for dense data, and include sample-size annotations. It also covers colorblind-safe palettes, multi-panel layout strategies, and exporting figures with controlled DPI and bounding boxes for downstream use.

When to use it

• Preparing figures for manuscripts, posters, or talks
• Diagnosing misleading plot shapes or unexpected distributions
• Designing multi-panel figures or stacked plots for temporal/category trends
• Ensuring color accessibility and consistent legend/sample-size reporting
• Exporting images to meet size/dpi constraints for journals or AI tools like Claude

Best practices

• Avoid KDE-based violin plots on log-scaled axes; either log-transform data first or use boxplots/histograms on log axes
• Always plot all data (show fliers) during initial QC; filter outliers only with documented technical reasons
• Tune element sizes for density: smaller markers (s ~ 20–40), thinner lines (linewidth ~ 1–1.5), and modest text sizes for small panels (fontsize 7–9)
• Place statistical annotations inside plot bounds using data-range calculations and explicit y-limits (e.g., y_pos = y_max * 0.9)
• Use colorblind-safe palettes (Okabe-Ito or Paul Tol) and keep colors consistent across panels; include sample sizes in legend labels
• Export with tight_layout/tight bbox, set dpi (150–300) and explicit figure size to meet publication and API image-size limits

Example use cases

• Replace distorted violin-on-log figures with log-transformed violin or boxplot to correctly show skewed distributions
• Create a dual-panel temporal figure: stacked area (proportions) + stacked bar (absolute counts) with per-category n in legends
• Diagnose unexpected smooth shapes: verify whether axis transforms were applied after density estimation
• Generate publication PNG/PDF with dpi and bbox_inches='tight' for journal submission or constrained API upload
• Annotate p-values and sample sizes clearly within multi-panel layouts to satisfy journal and reviewer expectations