academic-plotting

What it does

Generates publication-quality figures for ML/AI conference papers. For architecture diagrams, it uses Gemini to create architecture/system diagrams from paper context. For data figures, it auto-selects chart types (e.g., line plots, grouped bars, heatmaps) and generates data-driven figures using matplotlib/seaborn. It guides users to choose between Diagram figures (architecture/workflow) and Data figures (numerical results). It also prescribes applying publication styling, exporting outputs, and highlighting the paper's method in a distinct color.

How it works

• It reads the provided context (paper section, description, or results data) to extract visual entities and relationships for diagrams, or to extract dimensions, metrics, and comparisons for charts.
• For diagrams, it follows a workflow to identify components and connections, determine a layout pattern (left-to-right, layered, hub-and-spoke, etc.), assign colors per logical group, and write labels exactly from the source text. It uses Gemini via a prescribed prompt structure and a generation script template to render an architecture diagram image.
• For data charts, it reads the data, identifies axes and metrics, and chooses a chart type from a priority guide (e.g., line plot for time/steps, grouped bar for comparing categories, heatmap for grid data, stacked bar for part-of-a-whole). It then applies publication styling, highlights the method under study with a distinct color, and exports outputs in PDF and PNG formats. It provides a Python example showing how to structure the plotting code and styling.
• It requires specific dependencies (matplotlib>=3.8.0, seaborn>=0.13.0, numpy, google-genai>=1.0.0) and a GEMINI-related workflow for diagrams.

When to use it

Use when creating any figure for a conference paper, selecting between architecture/design diagrams (via Gemini) and data-driven figures (via matplotlib/seaborn).

What it can touch

• Diagram workflows imply using Gemini (via the prescribed API and prompt sections).
• Data workflows touch matplotlib and seaborn plotting code and associated data structures (CSV/JSON inputs as needed).

Caveats

• License is MIT; the skill declares dependencies and requires an environment variable for Gemini prompts. It emphasizes exact text labels from the paper and consistent styling, with guidance to export outputs at high resolution.
• The provided instructions require an API-based image generation process and a structured six-section Gemini prompt, along with a script template. It warns that quality may depend on multiple generation attempts and environmental setup (e.g., GEMINI_API_KEY).